feat: add schema-driven fal.ai external provider

[Xpos587]

Summary

Adds fal.ai as a schema-driven native external provider for InvokeAI.

• Adds dynamic fal.ai catalog discovery from /v1/models with cursor pagination and cross-page search.
• Reads each endpoint's queue OpenAPI schema and normalizes common inputs, output kind, and safe public properties.
• Installs image-capable endpoints as lightweight External Models with native Canvas routing.
• Keeps curated FLUX Schnell, FLUX Dev, FLUX Kontext Pro, and FLUX Fill presets.
• Adds schema-driven payload mapping for prompt, init/reference images, masks, size, seed, output count, and declared advanced JSON fields.
• Adds fal.ai Generic Media node for all other endpoint families: text/video, image-to-video, video-to-video, upscale, audio, speech, 3D, training, and future endpoints.
• Adds local image/mask/video upload placeholders for generic workflows and raw JSON output.
• Adds catalog UI with refresh, search, pagination, image-model Canvas install, and generic-workflow labels.
• Preserves source URL and dynamic panel controls in installed external model records.
• Documents model discovery, native Canvas boundaries, generic media workflows, and billing behavior.

Related Issues / Discussions

None.

QA Instructions

No billable fal.ai inference request was submitted. Provider tests mock catalog, schema, queue, CDN upload, polling, and result downloads.

Passed locally:

• 92 focused backend tests across fal provider/catalog, invocations, routers, model install, and existing external generation.
• 2,204 frontend tests via pnpm build.
• Frontend TypeScript check.
• Frontend ESLint, Prettier, dependency-cycle checks, and production build.
• Ruff check and format check.
• OpenAPI schema generation and frontend type generation.
• Native invocation registration check for fal_generic_media_native.

Manual verification after merge:

1. Configure fal.ai under Models -> Add Model -> External Providers.
2. Search the fal.ai catalog and install an image endpoint for Canvas.
3. Select it in Canvas; supported modes and common controls come from endpoint metadata.
4. Use fal.ai Generic Media in Workflow Editor for video/audio/3D/other endpoints, using the endpoint's JSON schema.

The repository checkout lacks Git LFS locally, so the unrelated full-suite model fixture tests cannot run in this environment. Fork PR checks are currently action_required until an InvokeAI maintainer approves fork workflow execution.

Merge Plan

No database migration. External model entries are weightless references. Existing curated external models and provider flows remain compatible.

Checklist

• Short, descriptive title
• Tests added / updated
• Documentation added / updated
• Updated What's New copy (not a release PR)