Provide async connect() variant for running event loop contexts

[beonde]

Problem

CapiscioMCPServer.connect() and the underlying MCPServerIdentity.from_env_sync() / connect_sync() use asyncio.run(), which raises RuntimeError: asyncio.run() cannot be called from a running event loop when invoked from:

• Jupyter notebooks
• ASGI frameworks (FastAPI, Starlette)
• FastMCP setups that already have an event loop
• Any async def entrypoint

This is a common usage pattern for MCP server bootstrapping code.

Proposed fix

• Add CapiscioMCPServer.connect_async() (or make connect() async and provide connect_sync())
• Alternatively, detect a running loop and use run_coroutine_threadsafe as other sync wrappers in the repo do
• Document the limitation prominently in the meantime

Affected methods

• CapiscioMCPServer.connect() → integrations/mcp.py:418
• MCPServerIdentity.from_env_sync() → connect.py:687
• MCPServerIdentity.connect_sync() → connect.py:687

Context

Identified during PR #32 code review (Copilot review comments on connect.py:687 and mcp.py:418).