RFC: MCP Optimizer migration to vMCP (#22)

* RFC: MCP Optimizer migration to vMCP

This RFC proposes migrating MCP-Optimizer functionality from a standalone
Python service to native vMCP tools, unifying the codebase and enabling
full integration with ToolHive's authentication, authorization, and
observability infrastructure.

Add two new optimizer tools as vMCP-native MCP tools:
- `optim.find_tool`: Semantic and keyword-based tool discovery using
  hybrid search (embeddings + BM25)
- `optim.call_tool`: Dynamic tool invocation on any backend server

These tools operate on vMCP's internal state and provide meta-capabilities
over aggregated backend tools.

- **Storage**: SQLite as the default embedding storage backend with a
  flexible interface for future providers
- **Embeddings**: HuggingFace Text Embeddings Inference (TEI) as the
  initial embedding provider, with interface supporting future vLLM/OpenAI
- **Authentication**: Full reuse of vMCP's two-boundary auth model with
  no modifications required
- **Scope**: One-to-one relationship between optimizer and vMCP instance
  (no cross-vMCP search)

* feedback by Jeremy

* Update based on Ozz feedback

Author: aponcedeleonch