fix(runtime): time out host_request when no host replies

[Electricitysheep]

Summary

host_request() awaits a Jupyter comm reply that never arrives when the module runs outside a running Prime Agent session (e.g. a bare kernel venv, or a test harness without a host). It hangs forever — no timeout, no error.

This is exactly the trap encountered while testing MCP integrations: importing test_mcp and calling list_tools() from a standalone kernel venv blocks indefinitely because McpIntegration._resolve_config() calls host_request("mcp.config", ...) and nothing replies.

Changes

prime-agent-runtime/src/rlm/__init__.py:

• Added HOST_REQUEST_TIMEOUT_SECONDS = 30.
• Wrapped the pending future in asyncio.wait_for(...); on timeout, cancel the future, close the comm, and raise a RuntimeError that explains the request requires a running Prime Agent session (instead of hanging).

prime-agent-runtime/test/test_host_request_timeout.py (new):

• Timeout path: with the timeout patched to 0.2s and a comm that never replies, host_request raises RuntimeError containing "timed out" and "Prime Agent host".
• Reply path: a comm that delivers a reply resolves normally (no spurious timeout).

Verification

All 66 tests in prime-agent-runtime/test pass (64 existing + 2 new), run with the source on PYTHONPATH against the kernel venv Python 3.11.

Note

Time out host_request after 30 seconds when no host replies

Previously, host_request would hang indefinitely if no host replied. It now wraps the response future in asyncio.wait_for with a 30-second timeout. On timeout, the pending future is canceled, the Comm is closed, and a RuntimeError is raised. A new test suite in test_host_request_timeout.py covers both the timeout and successful-reply paths.

^{Macroscope summarized 28ea5d9.}