chore(agentnetwork): fix config grouping, dead code, comments, coverage

- Move the new zone config key into the existing AgentNetwork config
  group (management/internals/server/config/config.go) instead of a
  sibling top-level field, wire modules.go to the new path, add the
  matching field to combined/cmd/config.go's AgentNetworkConfig and its
  mapping (it was previously unreachable in the combined binary), and
  document the key in infrastructure_files/management.json.tmpl and
  combined/config.yaml.example.

- Delete PickUnique and its three tests: Task 5 removed its last
  production caller, leaving it dead exported code with a stale
  words.go comment pointing at it.

- Reword two test comments that referenced our private review process
  instead of stating what the test locks down / why TargetId stays
  pinned to Cluster.

- Add TestBootstrapSettings_NonRetryableErrorFailsImmediately: a
  regression that dropped the isUniqueConstraintError gate and retried
  on every error would leave every existing allocator test green.

- Fix TestSynthesizeServiceForDomain_DegenerateInput's docstring: the
  early-return guard is an optimisation, not what makes "" and
  "localhost" resolve to no service.

- Replace manager.go's allocation-comment archaeology (a deleted
  per-cluster "taken" set, an `accountID[:4]` suffix, "~68 minutes") with
  the actual invariant, and note why each retry attempt gets its own
  transaction (a failed statement poisons the enclosing transaction on
  postgres).

- Collapse the per-service "no matching zone apex" debug log in
  account.go into a single line per call instead of one per skipped
  service.

- Document why idx_agent_network_settings_cluster_subdomain must stay
  on mysql: its index tag is what sizes subdomain as varchar(191)
  rather than longtext, which the new unique index requires.

combined/config.yaml.example

@@ -147,3 +147,11 @@ server:
   #   # is re-read periodically (mtime poll). An explicitly configured path that
   #   # fails to load fails startup; runtime reload errors keep the previous table.
   #   pricingDefaultsFile: "pricing.yaml"
+  #
+  #   # Parent DNS zone that Agent Network gateway endpoints are allocated
+  #   # under, producing <subdomain>.<zone>. Empty (the default) preserves the
+  #   # legacy behaviour of deriving the endpoint from the serving cluster, so
+  #   # self-hosted deployments are unaffected. Captured onto each settings row
+  #   # when that row is created; changing it later does not move existing
+  #   # tenants.
+  #   zone: "gateway.example.com"