Add Ollama embedding client

[theantichris]

Summary

Add an embedding client to the LLM package that calls Ollama's /api/embed endpoint. This is the foundation for the memory system's vector search.

Details

New files:

• internal/llm/embed.go
• internal/llm/embed_test.go

Types:

• EmbedRequest{Model, Input string}
• EmbedResponse{Embeddings [][]float64}

Function:

• Embed(ctx, host, model, input string) ([]float64, error) — POST to /api/embed using carlmjohnson/requests

Error:

• ErrEmbed = errors.New("neural embedding generation failed")

Testing:

• Table-driven tests with httptest.NewServer (same pattern as ollama_test.go)
• Cases: successful embedding, model not found (404), unexpected status, empty input

Notes:

• No changes to existing code
• Follow the same HTTP pattern used in Chat() and StreamChat() in ollama.go
• Use handleHTTPErrors() for consistent error handling

Acceptance Criteria

• Embed() function returns embedding vector from Ollama
• Proper error handling for 404 (model not found) and other HTTP errors
• Table-driven tests pass with mocked Ollama endpoint
• No changes to existing files