fix(mcp): reconnect after MCP call timeout

[bitboarderlabs]

Summary

• cancel MCP loop work when a configured tool-call timeout elapses
• treat MCP call timeouts as a stale transport signal: reconnect the long-lived MCP server task and retry once
• reuse the session-expired reconnect path for timeout recovery so circuit breakers are reset on successful reconnect

Tests

• /home/alex/.hermes/hermes-agent/venv/bin/python -m pytest tests/tools/test_mcp_tool_session_expired.py tests/tools/test_mcp_tool.py tests/tools/test_mcp_circuit_breaker.py tests/tools/test_mcp_reconnect_signal.py tests/tools/test_mcp_probe.py -q

[alt-glitch]

Related: #6665 (MCP timeout cancellation + session TOCTOU race). This PR extends the timeout handling to also reconnect the transport and retry, building on the same problem space.

[bitboarderlabs]

Rebased this PR onto current upstream main (3034eee) and resolved the conflict in tools/mcp_tool.py by keeping the new parallel-safe server registry plus the timeout/session reconnect helpers.\n\nLocal MCP-focused test run passed:\n\n\n/home/alex/.hermes/hermes-agent/venv/bin/python -m pytest tests/tools/test_mcp_tool_session_expired.py tests/tools/test_mcp_tool.py tests/tools/test_mcp_circuit_breaker.py tests/tools/test_mcp_reconnect_signal.py tests/tools/test_mcp_probe.py -q\n230 passed in 17.26s\n