feat(console): proxy remaining free Zen models and keep proxied 404 errors - #53963

Issue for this PR

N/A

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

Two changes to the legacy Zen → new inference proxy.

Proxied 404s were unreadable. When new inference answers 404 (for example an upstream provider no longer serving a model), SolidStart replaces the response with its own Cannot find any route matching [POST] …/zen/v1/chat/completions, so callers saw a routing error instead of the real one. Proxied 404s now return 400 with new inference's original body and headers, the same workaround the legacy handler already uses for its own provider responses. This applies to every proxied request, including migrated workspaces.

Last free models move to the direct proxy. big-pickle, muse-spark-1.3-contributor-free, and space-bunny-free join the models from #53717 and #53959. Those earlier batches serve keyless requests directly from new inference in production, so the remaining busy free models follow. deepseek-v4-flash-free (missing from the new registry) and mimo-v2.5-free (deprecated) stay on the legacy path.

How did you verify your code works?

bun typecheck in packages/console/app. After deploy, each model gets a keyless smoke request, and ling-3.0-flash-fin-free (whose OpenRouter :free variant currently has no providers) should return 400 with new inference's error body instead of SolidStart's route-not-found.

Screenshots / recordings

N/A

Checklist

• I have tested my changes locally (typecheck only; see above)
• I have not included unrelated changes in this PR