fix(google): remove the shut-down veo-3.1-generate-preview registration - #585

veo-3.1-generate-preview is the last Veo id in the Google videos catalog, and the Gemini API no longer serves it. The Veo 3.1 page says: "Veo 3.1 Preview (veo-3.1-generate-preview, veo-3.1-fast-generate-preview, and veo-3.1-lite-generate-preview) in the Gemini API was shut down on October 22, 2026. Migrate to Gemini Omni Flash (gemini-omni-1.2-flash) or the GA models available through the Gemini Enterprise Agent Platform to avoid service interruptions." The deprecations table lists veo-3.1-generate-preview with shutdown "October 22, 2026" and replacement gemini-omni-1.2-flash. On Google Cloud, the Veo 3.1 page lists only veo-3.1-generate-001, veo-3.1-fast-generate-001 and veo-3.1-lite-generate-001, not the preview id. CLAUDE.md: "Remove a model id only after a live authenticated call hard-rejects it". It does now (see Validation). #550 and #552 removed the other two Veo 3.1 preview ids the same way.

Fix:

• src/celeste/modalities/videos/providers/google/models.py: drop the veo-3.1-generate-preview entry, so GOOGLE_VEO_MODELS is empty.
• tests/unit_tests/test_google_videos_interactions.py: the Veo dispatch test patches _VEO_MODEL_IDS with a test id, so it still asserts that GoogleVideosClient selects GoogleVeoVideosClient and copies its endpoints. The Veo parse test builds GoogleVeoVideosClient directly with a test-local Model.
• tests/unit_tests/test_google_transport.py: the download credential-scoping test is parametrized over the two backend classes (GoogleVeoVideosClient, GoogleInteractionsVideosClient) with a test-local Model, instead of create_client with catalog ids. Both download paths stay covered.
• tests/unit_tests/test_vertex_routing.py is unchanged. It builds a test-local Model, the id there is only a URL fragment, and the file already uses ids that are no longer in the catalog (e.g. deepseek-chat, fix: provider bugs and catalog corrections from model-watch findings #314).

Left for the maintainer: the Veo route stays: veo.py, GOOGLE_VEO_PARAMETER_MAPPERS, GOOGLE_SUPPORTED_MIME_TYPES, the dispatcher branch and celeste.providers.google.veo. With GOOGLE_VEO_MODELS empty, no model id reaches that branch. An unregistered Veo id passed to create_client already raised "Unknown Google videos model" before this PR. The open decision is whether to keep Veo as a Google Cloud-only route, registering the GA ids above after a live Vertex call, or to delete the route.

Validation:

• make ci passes (763 unit tests).
• Live, Gemini API key: GoogleVideosClient.generate(prompt="A cat walking", duration=4, resolution="720p") on veo-3.1-generate-preview (strategy GoogleVeoVideosClient). POST /v1beta/models/veo-3.1-generate-preview:predictLongRunning returned 404 NOT_FOUND: "models/veo-3.1-generate-preview is not found for API version v1beta, or is not supported for predictLongRunning." GET /v1beta/models lists no Veo model, only gemini-omni-flash-preview, gemini-omni-1.1-flash and gemini-omni-1.2-flash.
• Google Cloud (GoogleADC): not proven live. The same call returned 403 because the local ADC project lacks aiplatform.endpoints.predict ("or it may not exist"). The Google Cloud page above does not list the id.