Add Gemini embedding support to config parser

[hd719]

Problem

The README documents Gemini as a supported embedding provider, but config.ts hardcodes provider: 'openai' in the parsed output and uses the openai SDK exclusively. The assertAllowedKeys on embedding config doesn't include provider as a valid key, so passing provider: 'gemini' would throw.

Current behavior

• config.ts type: provider: 'openai' (literal)
• assertAllowedKeys allows only ['apiKey', 'model'] — no provider key
• Embeddings class instantiates new OpenAI() unconditionally
• vectorDimsForModel only has OpenAI model dimensions

Proposal

1. Add 'provider' to allowed embedding keys
2. Expand the MemoryConfig type to provider: 'openai' | 'gemini'
3. Add a GeminiEmbeddings class (using @google/generative-ai or REST API)
4. Add Gemini model dimensions to EMBEDDING_DIMENSIONS (text-embedding-004: 768, embedding-001: 768)
5. Factory pattern in register() to pick the right embeddings client

Impact

Unblocks users who want free-tier Gemini embeddings as documented in the README.