fix(test): resolve thread leak failures in CI

dimos/core/test_native_module.py

@@ -18,8 +18,10 @@
 The echo script writes received CLI args to a temp file for assertions.
 """
 
+from collections.abc import Generator
 import json
 from pathlib import Path
+import threading
 import time
 
 import pytest
@@ -90,22 +92,35 @@ def start(self) -> None:
         pass
 
 
-def test_process_crash_triggers_stop() -> None:
-    """When the native process dies unexpectedly, the watchdog calls stop()."""
+@pytest.fixture
+def crash_module() -> Generator[StubNativeModule, None, None]:
+    """Create a StubNativeModule that dies after 0.2s, ensuring cleanup."""
+    before = {t.ident for t in threading.enumerate()}
     mod = StubNativeModule(die_after=0.2)
-    mod.pointcloud.transport = LCMTransport("/pc", PointCloud2)
-    mod.start()
+    yield mod
+    # The watchdog calls stop() from its own thread, which sets
+    # _module_closed=True. A second stop() from here is then a no-op,
+    # so we explicitly join any threads the test created.
+    for t in threading.enumerate():
+        if t.ident not in before and t is not threading.current_thread():
+            t.join(timeout=5)
+
+
+def test_process_crash_triggers_stop(crash_module: StubNativeModule) -> None:
+    """When the native process dies unexpectedly, the watchdog calls stop()."""
+    crash_module.pointcloud.transport = LCMTransport("/pc", PointCloud2)
+    crash_module.start()
 
-    assert mod._process is not None
-    pid = mod._process.pid
+    assert crash_module._process is not None
+    pid = crash_module._process.pid
 
     # Wait for the process to die and the watchdog to call stop()
     for _ in range(30):
         time.sleep(0.1)
-        if mod._process is None:
+        if crash_module._process is None:
             break
 
-    assert mod._process is None, f"Watchdog did not clean up after process {pid} died"
+    assert crash_module._process is None, f"Watchdog did not clean up after process {pid} died"
 
 
 @pytest.mark.slow