Add fastapi-mcp-client extension

[ulises-jeremias]

Problem

The proposal mentions MCP tools for chat and multi-agent apps. MCP is a reusable
capability that should not be duplicated across every AI extension.

CPA needs a focused MCP client extension that can be composed with FastAPI AI apps.

Solution

Create extensions/fastapi-mcp-client compatible with fastapi-backend.

Scope:

• Add MCP client configuration, for example mcp_servers.yaml or an env-driven
  equivalent.
• Add a typed tool registry/service that can be imported by chat or LangGraph code.
• Include one local/example MCP server only if it is lightweight and testable.
• Add tests with fake/local transports; no network dependency.
• Add docs:
  • template/docs/MCP_CLIENT_GUIDE.md
  • template/docs/README.md.append
  • .env.example.append if needed.
• Add partial pyproject.toml dependencies.

Constraints:

• Do not assume a specific LLM orchestration framework.
• Do not hardcode filesystem or database tools enabled by default.
• Document the security model: MCP servers are external capability providers and
  should be treated as trusted integrations.
• No GitHub Actions in this extension.

Acceptance criteria:

• Generated app has a deterministic test for listing/invoking a fake MCP tool.
• Docs show how to add a server without editing core chat code.

[ulises-jeremias]

This item is part of the MLOps roadmap. If you'd like to work on it, comment here.