feat(embed): add local embedding providers

[GaosCode]

Summary

Adds the embedding provider layer for OpenAI-compatible services, Ollama, and llama.cpp without implementing vector storage or semantic search.

Changes

• Adds internal/embed provider abstraction and HTTP clients
• Adds provider-specific defaults for OpenAI, Ollama, llama.cpp, and OpenAI-compatible endpoints
• Extends embedding config with base_url, max_input_chars, and request_timeout
• Adds non-fatal doctor checks for enabled embedding providers

Testing

• go test ./...

Risks / Notes

• This does not drain embedding_jobs, add message_embeddings, or implement semantic/hybrid search.
• Made with Codex