fix(api): stop leaking the internal provider UUID in /v1/models and honor the configured prefix (#8327)

[diegosouzapw]

Refs #8327

Root cause

src/app/api/v1/models/catalog.ts builds two different identifiers per model
entry for compatible-provider nodes (openai-compatible / anthropic-compatible):

• alias — DOES receive providerIdToPrefix (built from provider nodes) and
  correctly becomes the operator's configured prefix for the published id field.
• canonicalProviderId (via resolveCanonicalProviderId()) — used for owned_by,
  but only knows the static AI_PROVIDERS/PROVIDER_MODELS alias maps. A
  compatible-provider node's id is an internal UUID (e.g.
  openai-compatible-chat-<uuid>, per createProviderNode()'s
  id: data.id || uuidv4()) that is never present in those static maps, so it
  falls through every lookup and returns the raw UUID verbatim — leaking the
  internal provider-node id into the public owned_by field for every emit
  site that uses it: the synced-models block (the one matching the reporter's
  scenario), the custom-models block, the model-alias block, and the
  managed-fallback block (which additionally skipped canonicalProviderId
  resolution entirely, using the raw providerId directly).

Fix

canonicalProviderId is still required, unmodified, by internal
connection/hidden-model/registry lookups that are keyed on the raw provider-node
id (getConnectionsForProvider, getModelIsHidden, etc.) — so I did not change
what it resolves to. Instead I added a dedicated resolvePublicOwnerId(providerId, canonicalProviderId) helper that checks providerIdToPrefix first and applied
it at every owned_by emit site (previously owned_by: canonicalProviderId /
owned_by: providerId). Built-in providers are unaffected — providerIdToPrefix
only has entries for actual provider nodes with a configured prefix, never for
static AI_PROVIDERS ids.

Scope

This PR fixes symptom (a) — the internal provider UUID leaking as owned_by —
and symptom (b) — the configured prefix being ignored for owned_by. It does
not touch effort-variant multiplication, which the triage plan confirmed is
intentional, documented design (open-sse/utils/claudeEffortVariants.ts,
open-sse/utils/syncedEffortVariants.ts), not a defect. Using Refs #8327
rather than Closes #8327 since the issue as filed also raises that symptom.

Regression test (Hard Rule #18 — TDD)

New file tests/unit/8327-models-owned-by-prefix.test.ts, reusing the triage
plan's proven repro scenario (compatible provider node + configured prefix +
synced/custom models).

• RED confirmed against the pre-fix code (temporarily reverted
  catalog.ts via git checkout HEAD -- <file> in-worktree, no stash used):

  AssertionError [ERR_ASSERTION]: owned_by must be the configured prefix
  "pix4k-talk", not the raw provider-node id — got
  "openai-compatible-chat-550e8400-e29b-41d4-a716-446655440000"
  

• GREEN after restoring the fix: all 3 new tests pass (synced models, custom
  models, and a built-in-provider contract-preservation check).

Gates run (all green)

• node scripts/check/check-file-size.mjs — OK
• node scripts/check/check-complexity.mjs — pre-existing baseline drift
  (2167 violations), confirmed identical on a disposable probe worktree pinned
  to origin/release/v3.8.49 — not introduced by this diff.
• node scripts/check/check-cognitive-complexity.mjs — same pre-existing
  drift (956), confirmed identical on the same probe — not introduced by this diff.
• npm run typecheck:core — clean
• npx eslint --suppressions-location config/quality/eslint-suppressions.json <changed files> — 0 errors
• Sibling tests of the touched area (26 files covering api/v1/models/catalog) — all green, including models-catalog-route.test.ts (43 tests)
• node scripts/check/check-changelog-integrity.mjs — OK

Changelog

changelog.d/fixes/8327-models-owned-by-prefix.md