fix(embeddings): forward output dimensions to Gemini for consistent embedding dims

[diegosouzapw]

Summary

Forward the OpenAI-style dimensions request field to the Gemini-native
outputDimensionality field so clients can reliably get the embedding
vector size they ask for.

Ported from upstream PR decolua/9router#1366 (author @nguyenha935).

Why

Gemini embedding models (text-embedding-004, gemini-embedding-001,
gemini-embedding-2-preview) default to 3072-dim vectors. OpenAI-compatible
clients typically request a smaller size (e.g. 1536 for pgvector schemas)
via the OpenAI dimensions field. OmniRoute already forwarded dimensions to
Gemini's OpenAI-compatibility endpoint (/v1beta/openai/embeddings), but
Google's compatibility shim does not document the dimensions →
outputDimensionality translation — so callers still received full-size
vectors back. The upstream PR fixes the same gap on the native :embedContent
/ :batchEmbedContents path; this port applies the equivalent fix to
OmniRoute's OpenAI-compat embeddings handler.

What changed

• open-sse/handlers/embeddings.ts — when provider === "gemini" and the
  client sent a positive finite dimensions, mirror it into the
  outputDimensionality field of the upstream body (alongside the existing
  dimensions forwarding). 0, NaN, negative values are ignored. Other
  providers are unchanged.
• tests/unit/embeddings-gemini-dimensions.test.ts — 5 new tests covering:
  Gemini single input, Gemini batch input, omitted dimensions (no injection),
  non-Gemini providers (no leak), invalid dimensions (0 rejected).

Test plan

• node --import tsx/esm --test tests/unit/embeddings-gemini-dimensions.test.ts → 5/5 pass (RED→GREEN confirmed)
• node --import tsx/esm --test tests/unit/embeddings-nvidia-input-type.test.ts tests/unit/embeddings-handler.test.ts tests/unit/embeddings-auth.test.ts → 24/24 pass (no regression)
• npx eslint open-sse/handlers/embeddings.ts tests/unit/embeddings-gemini-dimensions.test.ts → clean
• Husky pre-commit + pre-push hooks → pass

Attribution

Original author: @nguyenha935 (upstream decolua/9router#1366).

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.