[management,proxy] Agent network: per-account LLM gateway (policy, metering, multi-provider) (#6555)

* [agent-network] Shared proto, OpenAPI schema, and generated types

* [agent-network] Management: store, manager, synthesizer, policy engine, provider catalog, HTTP/gRPC API

Adds the account-scoped agent-network module: provider/policy/budget CRUD and
store, the reverse-proxy service synthesizer, policy selection + limit
enforcement, the provider catalog (incl. Vertex AI and AWS Bedrock entries),
and the management HTTP + proxy gRPC surfaces.

* [management] Fix agent-network proxy-peer fan-out on affected-peer recompute

The affected-peers resolver loaded only persisted reverse-proxy services, but
agent-network services are synthesized on demand and never persisted. As a
result the embedded proxy peer was never folded into the affected set when a
client's group changed, so the proxy received no network-map update for a newly
authorised client and rejected its handshake until a full resync (restart).

loadProxyServices now merges the synthesized agent-network services (injected
via a registration hook to avoid an import cycle), so proxy peers learn newly
authorised clients immediately.

* [proxy] Reverse-proxy middleware framework, chain, and request plumbing

The per-target middleware chain (slots, dispatcher, mutation gate, metadata
merger), body capture, access-log terminal sink, and the proxy wiring that
builds + runs chains for synthesized agent-network services.

* [proxy] LLM parsers, pricing, and builtin middlewares (OpenAI, Anthropic, Vertex AI, AWS Bedrock)

Request/response parsers and SSE/event-stream metering, the embedded pricing
table, and the builtin middleware set: request parser, router, policy
limit-check/record, cost meter, guardrail, identity inject, response parser.
Includes the path-routed providers — Google Vertex AI (keyfile:: service-account
OAuth minting) and AWS Bedrock (bearer auth, invoke/converse/streaming, optional
/bedrock prefix) — plus the Models allowlist and unmeterable-publisher deny.

* [proxy] IPv6 in-place apply and TCP accept-loop hardening on netstack listeners

* [agent-network] End-to-end test suite, module docs, and deployment preset

* [agent-network] Fix codespell typos and exclude false positives

- labelgen word pool: vermillion -> vermilion, racoon -> raccoon.
- codespell ignore list: add flate (Go compress/flate package), recordin
  (a test-local identifier), and unparseable (a valid alternative spelling used
  consistently across identifiers + a metadata-value constant).

* [management] Set LastSeen on injected proxy peer in realstack test (MySQL strict-mode)

The injected embedded proxy peer had a PeerStatus with a zero LastSeen, which
serializes to '0000-00-00' and is rejected by MySQL in strict mode (SQLite
tolerates it). Set LastSeen to a valid time so SaveAccount succeeds on both
engines.

* [agent-network] Remove e2e shell-script suite from this branch

The end-to-end shell scripts under scripts/e2e/ are maintained in a separate
testing suite and are not part of this change set.

* [agent-network] Polish module docs: remove internal review scaffolding, fix links, verify diagrams

Strip PR-review framing, commit references, absolute paths, and stale internal
references from the agent-network module docs; fix broken relative links; verify
all diagrams against the current architecture. Remove the internal AI-reviewer
prompt file.

* [management] Refine session expiration handling to support 3-state encoding for SSO deadlines

* [agent-network] Relocate agentnetwork package to internals/modules

Move management/server/agentnetwork (and its catalog/, labelgen/, types/
subpackages) to management/internals/modules/agentnetwork, alongside the
reverse-proxy module, and rewrite all importers. Pure relocation: package names,
the synthesizer + affectedpeers registration hook, and store access (shared
store.Store) are unchanged, so no import cycle is introduced (affectedpeers
still depends only on the agentnetwork/types leaf).

* [agent-network] Co-locate HTTP handlers in the module (RegisterEndpoints)

Move the agent-network HTTP handlers from server/http/handlers/agentnetwork into
the module at internals/modules/agentnetwork/handlers (package handlers) and
rename the entrypoint AddEndpoints -> RegisterEndpoints, matching the
reverse-proxy module convention. Wiring in http/handler.go updated accordingly.

proxy/middleware_translate.go

@@ -0,0 +1,165 @@
+package proxy
+
+import (
+	"context"
+	"time"
+
+	log "github.com/sirupsen/logrus"
+
+	"github.com/netbirdio/netbird/proxy/internal/middleware"
+	"github.com/netbirdio/netbird/proxy/internal/middleware/bodytap"
+	"github.com/netbirdio/netbird/shared/management/proto"
+)
+
+// translateMiddlewareCaptureConfig builds the per-target capture
+// limits used by the middleware chain. Returns nil when the options
+// are nil or no capture field is set. Negative caps are normalised to
+// zero; oversized caps are clamped to middleware.MaxBodyCapBytes.
+func translateMiddlewareCaptureConfig(targetID string, opts *proto.PathTargetOptions) *bodytap.Config {
+	if opts == nil {
+		return nil
+	}
+	reqCap := clampMiddlewareCaptureBytes(targetID, "request", opts.GetCaptureMaxRequestBytes())
+	respCap := clampMiddlewareCaptureBytes(targetID, "response", opts.GetCaptureMaxResponseBytes())
+	types := opts.GetCaptureContentTypes()
+	if reqCap == 0 && respCap == 0 && len(types) == 0 {
+		return nil
+	}
+	return &bodytap.Config{
+		MaxRequestBytes:  reqCap,
+		MaxResponseBytes: respCap,
+		ContentTypes:     types,
+	}
+}
+
+func clampMiddlewareCaptureBytes(targetID, direction string, v int64) int64 {
+	if v < 0 {
+		log.Debugf("target %s %s capture cap %d clamped to 0", targetID, direction, v)
+		return 0
+	}
+	if v > middleware.MaxBodyCapBytes {
+		log.Debugf("target %s %s capture cap %d clamped to %d", targetID, direction, v, middleware.MaxBodyCapBytes)
+		return middleware.MaxBodyCapBytes
+	}
+	return v
+}
+
+// translateMiddlewareConfigs converts the proto MiddlewareConfig list
+// into validated middleware.Spec values. The list is truncated to
+// middleware.MaxMiddlewaresPerChain when the caller exceeds the cap.
+// Entries with empty IDs, unknown IDs (when registry is non-nil), or
+// unspecified slots are skipped with a warn log. Timeouts are clamped
+// to [MinTimeout, MaxTimeout] and zero substitutes for DefaultTimeout.
+// Returns nil when the resulting slice is empty so callers can leave
+// PathTarget.Middlewares unset.
+func translateMiddlewareConfigs(
+	ctx context.Context,
+	targetID string,
+	in []*proto.MiddlewareConfig,
+	registry *middleware.Registry,
+) []middleware.Spec {
+	_ = ctx
+	if len(in) == 0 {
+		return nil
+	}
+	if len(in) > middleware.MaxMiddlewaresPerChain {
+		log.Warnf("middleware list for target %q truncated: %d entries exceeds cap of %d",
+			targetID, len(in), middleware.MaxMiddlewaresPerChain)
+		in = in[:middleware.MaxMiddlewaresPerChain]
+	}
+
+	out := make([]middleware.Spec, 0, len(in))
+	for _, cfg := range in {
+		spec, ok := translateMiddlewareConfig(targetID, cfg, registry)
+		if !ok {
+			continue
+		}
+		out = append(out, spec)
+	}
+	if len(out) == 0 {
+		return nil
+	}
+	return out
+}
+
+// translateMiddlewareConfig validates and converts a single
+// MiddlewareConfig. The second return value is false when the entry
+// must be dropped from the chain.
+func translateMiddlewareConfig(targetID string, cfg *proto.MiddlewareConfig, registry *middleware.Registry) (middleware.Spec, bool) {
+	if cfg == nil {
+		return middleware.Spec{}, false
+	}
+	id := cfg.GetId()
+	if id == "" {
+		log.Warnf("middleware config for target %q dropped: empty middleware id", targetID)
+		return middleware.Spec{}, false
+	}
+	if registry != nil && !registry.IsKnown(id) {
+		log.Warnf("unknown middleware %q configured for target %s; dropping", id, targetID)
+		return middleware.Spec{}, false
+	}
+	slot, ok := protoToMiddlewareSlot(cfg.GetSlot())
+	if !ok {
+		log.Warnf("middleware %q on target %q dropped: slot is unspecified", id, targetID)
+		return middleware.Spec{}, false
+	}
+
+	var rawConfig []byte
+	if src := cfg.GetConfigJson(); len(src) > 0 {
+		rawConfig = append([]byte(nil), src...)
+	}
+
+	return middleware.Spec{
+		ID:        id,
+		Slot:      slot,
+		Enabled:   cfg.GetEnabled(),
+		FailMode:  protoToMiddlewareFailMode(cfg.GetFailMode()),
+		Timeout:   clampMiddlewareTimeout(id, cfg.GetTimeout().AsDuration()),
+		RawConfig: rawConfig,
+		CanMutate: cfg.GetCanMutate(),
+	}, true
+}
+
+// protoToMiddlewareSlot maps the proto slot enum onto the internal
+// middleware.Slot. Returns ok=false for the UNSPECIFIED value so the
+// translator can drop the entry.
+func protoToMiddlewareSlot(s proto.MiddlewareSlot) (middleware.Slot, bool) {
+	switch s {
+	case proto.MiddlewareSlot_MIDDLEWARE_SLOT_ON_REQUEST:
+		return middleware.SlotOnRequest, true
+	case proto.MiddlewareSlot_MIDDLEWARE_SLOT_ON_RESPONSE:
+		return middleware.SlotOnResponse, true
+	case proto.MiddlewareSlot_MIDDLEWARE_SLOT_TERMINAL:
+		return middleware.SlotTerminal, true
+	default:
+		return 0, false
+	}
+}
+
+// protoToMiddlewareFailMode maps the proto FailMode enum onto the
+// internal middleware.FailMode, defaulting to FailOpen for any value
+// other than FAIL_CLOSED.
+func protoToMiddlewareFailMode(m proto.MiddlewareConfig_FailMode) middleware.FailMode {
+	if m == proto.MiddlewareConfig_FAIL_CLOSED {
+		return middleware.FailClosed
+	}
+	return middleware.FailOpen
+}
+
+// clampMiddlewareTimeout enforces the proxy-wide [MinTimeout, MaxTimeout]
+// bounds and substitutes DefaultTimeout for zero inputs. A warn is logged
+// only on an actual clamp, not when filling the default.
+func clampMiddlewareTimeout(id string, d time.Duration) time.Duration {
+	if d <= 0 {
+		return middleware.DefaultTimeout
+	}
+	if d < middleware.MinTimeout {
+		log.Debugf("middleware %s timeout %s clamped to %s", id, d, middleware.MinTimeout)
+		return middleware.MinTimeout
+	}
+	if d > middleware.MaxTimeout {
+		log.Debugf("middleware %s timeout %s clamped to %s", id, d, middleware.MaxTimeout)
+		return middleware.MaxTimeout
+	}
+	return d
+}