fix(provider/google-vertex): support embedding model settings

[Kage18]

Summary

Fixes #7262.

vertex.textEmbeddingModel() is documented with an optional settings argument such as { outputDimensionality }, but the provider type and constructor path only accepted a model id.

This adds optional default embedding settings to:

• vertex.embeddingModel(modelId, settings)
• vertex.textEmbeddingModel(modelId, settings)

The defaults are merged into embedding requests, and per-call provider options still take precedence.

Tests

• pnpm --dir packages/google-vertex exec vitest --config vitest.node.config.js --run src/google-vertex-provider-base.test.ts src/google-vertex-embedding-model.test.ts
• pnpm --filter @ai-sdk/google-vertex type-check

[Kage18]

Kage usage for this PR:

This PR fixed #7262 by allowing default embedding settings on vertex.embeddingModel(modelId, settings) and vertex.textEmbeddingModel(modelId, settings).

How Kage helped:

• Scoped recall to packages/google-vertex provider construction and embedding-model request behavior.
• Kept the relevant files together: google-vertex-provider-base.ts, google-vertex-embedding-model.ts, and their focused tests.
• Helped preserve the provider-option precedence rule: constructor defaults are merged into embedding requests, but per-call provider options still win.
• Captured that docs already implied settings such as outputDimensionality, while the provider type/constructor path did not support them.
• Captured the verification commands for provider-base tests, embedding-model tests, and package type-checking.

Why this matters for future sessions:
A future agent touching Vertex embeddings can recall the default-settings merge behavior and avoid breaking per-call overrides while changing provider construction.

Full contribution summary and Kage metrics: kage-core/Kage#8