feat(core): add Gemini Embedding 2 support

[Showiix]

Summary

• add gemini-embedding-2 metadata while keeping gemini-embedding-001 as the Gemini default
• handle gemini-embedding-2 batch requests by embedding each input separately, preserving one vector per input for indexing
• keep the existing batched request behavior for gemini-embedding-001
• add Gemini provider regression tests and update docs to mention gemini-embedding-2

Tests

• pnpm --filter @zilliz/claude-context-core test
• pnpm --filter @zilliz/claude-context-core typecheck
• pnpm --filter @zilliz/claude-context-core build

Fixes #365

[zc277584121]

Thanks for adding this. The model support itself is useful, but I think the batch behavior needs one change before we merge this.

I tested the current Google GenAI SDK with a real Gemini API key, and gemini-embedding-2 accepts array input through models.embedContent and returns one embedding per input:

gemini-embedding-2 single ok embeddings=1 dim=768
gemini-embedding-2 array ok embeddings=2 dims=768,768

Because of that, we should not mark gemini-embedding-2 as supportsBatchEmbedding: false or force it to embed each chunk one by one. That would make large codebase indexing much slower by increasing the number of API calls.

Could you update this PR to keep the existing batched request path for gemini-embedding-2, and limit the change to adding the model metadata, docs, and regression tests? The response-count validation and empty-batch handling are good additions and can stay.

[Showiix]

Thanks, updated the PR.