Fal category is on the model row, not an OpenAPI selector #561

Fal publishes a category on each model row (providerMetadata.category). It is not a selector on GET /v1/openapi/fal.

GET /v1/models?provider=fal (1453 models, fetched 2026-11-04) includes:

The OpenAPI operation getProviderOpenApi accepts model and activity only. There is no category parameter.

?activity= is the closest selector, and it still 400s:

• GET /v1/openapi/fal?activity=image — spec_requires_selector, model-grained, 638 endpoints
• GET /v1/openapi/fal?activity=video — same, 653 endpoints

The only 200 is one model. GET /v1/openapi/fal?model=fal-ai/flux/dev is two paths on https://queue.fal.run:

• POST /fal-ai/flux/dev
• GET /fal-ai/flux/dev/requests/{request_id}

TanStack AI's fal client is one queue client for every model: POST https://queue.fal.run/{endpointId} and GET https://queue.fal.run/{endpointId}/requests/{requestId}. Generating one client per model (1495 of them) does not match that client. A category document would also still hit the 40-endpoint cap: the smallest large categories above are 41 to 382 endpoints.

Same model-grained refusal, no category field at all:

• cloudflare-ai-gateway — 197 endpoints
• reactor — 145 endpoints

Ask

1. Accept ?category= on GET /v1/openapi/fal, using the published providerMetadata.category.
2. Publish one fal queue document: model id is a path parameter, and the body schema stays the per-model schema. That is the client we can generate once.
3. Category documents still need the 40-endpoint cap raised, or they 400 as well.