fix(model-menu): resolve per-chat backend so /model honours overrides

[claudiusthebot]

Summary

The /model command and its callbacks read gateway.backend (the
global chat-role default) instead of the per-chat backend, which
broke two things once per-chat overrides shipped via the pool
refactor:

1. A chat that switched to openai-agents for OpenRouter saw the
   Claude catalog instead of OpenRouter's — the menu was always
   rendered against the global default.
2. The free-only toggle (gated on freeCount > 0) never appeared
   because Claude's catalog was queried, not the active backend's.

Plus fetchEndpointModels was fire-and-forget, so the first /model
render after a backend switch could race the /models HTTP call
and show an empty catalog.

What changed

Per-chat backend resolution

• core/backend-controller.ts — new resolveChatBackend(chatId, fallback)
  helper: pool-first / gateway-fallback / null when neither is
  wired. Frontend-agnostic.
• frontend/telegram/model-menu.ts (NEW, ~260 LOC) — controller that
  consolidates main-menu, browse, and backend-submenu flows. Always
  resolves per-chat backend. Pure-function shape; no Telegram types
  leak in.
• frontend/telegram/commands.ts — /model, /reset warmSession,
  /status enrichment all switch to per-chat backend via the
  controller / resolveBackendForChat.
• frontend/telegram/callbacks.ts — every /model callback branch
  re-renders against the per-chat backend. Critical: after
  backend-select rebind, the menu re-render correctly uses the
  new backend's catalog instead of the stale closure-captured one.
• frontend/discord/{commands,callbacks}.ts — same treatment for
  /model, /settings, /status, and warmSession.

OpenRouter discovery awaitability

• backend/openai-agents/discovery.ts (NEW, ~248 LOC) — extracted
  endpoint-model discovery. startDiscovery stashes the in-flight
  promise on state; awaitDiscovery(timeoutMs) lets the picker wait
  briefly on first render so the catalog isn't empty just because the
  HTTP call hasn't returned. refreshDiscovery for explicit retries.
  Free-pricing detection is now string-or-number tolerant (vLLM ships
  numeric 0, OpenRouter ships string "0", "0.0", etc.).
• backend/openai-agents/init.ts — delegates to discovery.ts;
  caches baseURL/apiKey on state so triggerDiscoveryRefresh
  doesn't need to re-read config.
• backend/openai-agents/state.ts — adds discoveryPromise,
  discoveryAt, baseURL, apiKey fields.
• backend/openai-agents/models.ts — getSettingsPresentation is
  now async and awaits in-flight discovery before snapshotting.
  Picker first-render is no longer empty.

Result

• Backend submenu → OpenAI Agents → /model now shows the OpenRouter
  catalog (350+ models, grouped by provider).
• Free-only toggle appears only when the per-chat backend reports
  free models (so Claude/Codex: no toggle; openai-agents+OpenRouter:
  yes; pure OpenAI: no — provider has no free tier).
• First /model after a backend switch waits (up to 3s) for the
  catalog fetch rather than rendering empty.

Test plan

• 2587/2599 tests pass (109 in the touched areas), TypeScript
  clean, prettier clean, oxlint 12 warnings (down from 13 baseline).
• New: 22 discovery tests, 17 model-menu-controller tests.
• Existing: openai-agents-enrichment.test.ts import updated to
  discovery.ts; openai-agents-models.test.ts
getSettingsPresentation calls now await-ed.
• Live smoke: set enabledBackends: ["claude", "openai-agents"]
  with openaiBaseUrl: "https://openrouter.ai/api/v1", switch
  backend on a chat, /model should show OpenRouter catalog + Free
  toggle. Switch back to Claude: Free toggle disappears, Claude
  models shown.

🤖 Generated with Claude Code

[claudiusthebot]

Folded into #211 — both PRs were the same multi-backend work artificially split, per Dylan's ask to keep it as a single PR. The two commits from this branch (699d0be model-menu per-chat resolution + b16433b MCP tool-name dedupe) are now sitting on top of feat/multi-backend-hotswap after a clean fast-forward. PR #211's description has been updated to reflect the combined scope.

Closing.