Part of #1592

Scope

Implement the embedContent endpoint for Gemini's embedding models.

Files to Modify

• crates/zeph-llm/src/gemini.rs -- implement embed(), update supports_embeddings()
• crates/zeph-core/src/config/types.rs -- add embedding_model field to GeminiConfig

Key Implementation Details

• Endpoint: POST /v1beta/models/{model}:embedContent?key={API_KEY}
• Default model: text-embedding-004 (768 dimensions)
• Request includes taskType for optimal embedding:
  • RETRIEVAL_QUERY for search queries (used in embed() calls)
  • RETRIEVAL_DOCUMENT for stored documents
• Response: embedding.values as Vec<f32>
• supports_embeddings() returns true only when embedding_model is configured

Config

[llm.gemini]
model = "gemini-2.5-flash"
max_output_tokens = 8192
embedding_model = "text-embedding-004"

Acceptance Criteria

• supports_embeddings() returns true when embedding model configured
• embed() returns correctly dimensioned vectors
• taskType set to RETRIEVAL_QUERY for embed() calls
• Embedding model configurable in [llm.gemini]
• Compatible with existing Qdrant/semantic memory pipeline
• Unit tests with wiremock