Add OVHcloud embeddings usage extractor

[dsfaccini]

David's AICA here:

Summary

Flavor-coverage gap — same class as MiniMax #437.

The gateway's OVHcloud provider sets api_flavor: 'embeddings' for embeddings requests (providers/ovhcloud.ts → initializeAPIFlavor), but prices/providers/ovhcloud.yml only defined a chat extractor. Any embeddings call would cause extractUsage to throw:

Unknown apiFlavor 'embeddings', allowed values: chat

…tripping the gateway's circuit-breaker and returning invalid response, unable to extract usage to the caller.

Fix: add a second extractor for api_flavor: embeddings using prompt_tokens → input_tokens — the standard OpenAI-compatible embeddings usage shape.

- api_flavor: embeddings
  root: usage
  model_path: model
  mappings:
    - path: prompt_tokens
      dest: input_tokens

This matches the shape used by openai.yml and doubleword.yml embeddings extractors. OVHcloud AI Endpoints (oai.endpoints.kepler.ai.cloud.ovh.net) are fully OpenAI-compatible.

Verification status

Shape is mirrored from the OpenAI-compatible embeddings convention (openai.yml / doubleword.yml embeddings extractors). No OVHcloud API key was available to confirm the exact usage object empirically, and OVHcloud is not in the gateway's current live route list, so this fixes a latent gap. High confidence on shape (OpenAI-compatible); not empirically confirmed against OVHcloud.

Changes

• prices/providers/ovhcloud.yml — adds embeddings extractor (existing chat extractor unchanged)
• packages/js/src/data.ts, packages/python/genai_prices/data.py, prices/data.json, prices/data_slim.json — regenerated
• tests/dataset/usages.json — regenerated snapshot (no OVHcloud embeddings bodies in corpus; ordering of groq/mistral entries shifted alphabetically in unrelated bodies)
• packages/js/src/__tests__/extractUsage.test.ts — adds OVHcloud embeddings extraction test

[cubic-dev-ai]

No issues found across 7 files

_{Re-trigger cubic}