Bug: `process_is_alive` crashes Yoke runtime deployments on Windows (OSError: [WinError 87])

[Adityakk9031]

Describe the bug

The process_is_alive function in src/yoke/providers/runtime_deployment.py crashes on Windows when attempting to check if a process is still running during the cleanup of stale temporary runtime deployments.

When os.kill(pid, 0) is called on a non-existent PID on Windows, it raises a generic OSError: [WinError 87] The parameter is incorrect, rather than the ProcessLookupError seen on Unix systems. Because this exception is unhandled, it bubbles up and completely crashes the deploy_runtime function, preventing the harness from starting.

Steps to reproduce

1. Run Yoke on a Windows machine.
2. Have a stale/abandoned Yoke deployment directory (e.g. yoke-codex-<pid>-...) in your temporary folder, where the <pid> no longer exists.
3. Attempt to run any agent via Yoke.
4. The application crashes during reclaim_stale_deployments.

Expected behavior

The function should safely catch the Windows-specific OSError and return False, correctly identifying the process as dead and allowing the stale directory to be cleaned up without crashing.

Suggested Fix

Update process_is_alive in src/yoke/providers/runtime_deployment.py to catch OSError:

def process_is_alive(pid: int) -> bool:
    if pid == os.getpid():
        return True
    try:
        os.kill(pid, 0)
    except ProcessLookupError:
        return False
    except PermissionError:
        return True
    except OSError:
        # On Windows, os.kill(pid, 0) for a non-existent process raises OSError: [WinError 87]
        return False
    return True