Support VertexAI embedding task_type and title

[goianiense]

Summary

• Plumbs provider-specific kwargs (**provider_params) through RubyLLM.embed → Provider#embed → Protocol#embed → render_embedding_payload
• VertexAI embedding payloads can now include task_type (e.g. SEMANTIC_SIMILARITY, RETRIEVAL_QUERY, RETRIEVAL_DOCUMENT) and optional title on each instance
• Other embedding providers ignore unknown kwargs via ** splat so extra options don't raise

Fixes #810

Usage

RubyLLM.embed(
  "hello",
  model: "gemini-embedding-001",
  provider: :vertexai,
  dimensions: 1536,
  task_type: "SEMANTIC_SIMILARITY"
)

Produces a VertexAI payload shaped like:

{
  "instances": [
    {
      "content": "hello",
      "task_type": "SEMANTIC_SIMILARITY"
    }
  ],
  "parameters": {
    "outputDimensionality": 1536
  }
}

For document embeddings:

RubyLLM.embed(
  "document body",
  model: "gemini-embedding-001",
  provider: :vertexai,
  task_type: "RETRIEVAL_DOCUMENT",
  title: "My Document"
)

Test plan

• Added unit specs for VertexAI::Embeddings#render_embedding_payload covering task_type, title, batch instances, and dimensions
• Added specs proving RubyLLM.embed forwards task_type/title to the provider (reproduced ArgumentError: unknown keyword: :task_type before the fix)
• All new specs pass locally

[crmne]

Thanks for the PR.

The underlying Vertex AI embedding issue is fixed on main, but through the general one-shot params API rather than provider-specific top-level keywords. The fix landed in 1bc6fc0, with the follow-up keyword cleanup in 5cdf01f.

Use params for Vertex AI request fields:

RubyLLM.embed(
  "hello",
  model: "gemini-embedding-001",
  provider: :vertexai,
  dimensions: 1536,
  params: { task_type: "SEMANTIC_SIMILARITY" }
)

For document embeddings, params can include both task_type and title. RubyLLM applies those fields to each Vertex AI embedding instance.

Closing this PR as superseded by the implementation on main.