[pantafive]

Fixes #1140

The atexit handler added in #1017 runs Reaper.delete_instance(), which force-removes the Ryuk container while the interpreter shuts down. Ryuk reaps a session only RYUK_RECONNECTION_TIMEOUT after its connection drops, so a process that exits normally leaves every container of its session running.

The handler now closes only the socket. That still prevents the ResourceWarning from #1017, and Ryuk reaps the session after the timeout. delete_instance() keeps its behaviour for explicit callers and reuses the new close_socket().

     @classmethod
-    def delete_instance(cls) -> None:
+    def close_socket(cls) -> None:
         if Reaper._socket is not None:
             Reaper._socket.close()
             Reaper._socket = None

+    @classmethod
+    def delete_instance(cls) -> None:
+        Reaper.close_socket()
+
 ...
-        atexit.register(Reaper.delete_instance)
+        # Only close the socket: Ryuk reaps the session after RYUK_RECONNECTION_TIMEOUT unless it is removed first.
+        atexit.register(Reaper.close_socket)

test_reaper_removes_session_containers_after_interpreter_exit starts a container in a child interpreter with RYUK_RECONNECTION_TIMEOUT=1s, lets it exit normally and waits up to 30s for the container to disappear. It carries the same skipif(is_mac()) and inside_docker_check marks as test_wait_for_reaper.

Testing, on Colima (Linux VM) with the macOS skip lifted locally:

• before the fix: TimeoutError: Container 661e2b17… was not removed within 30s (status=exited)
• after the fix: all of tests/core/test_ryuk.py passes
• no ResourceWarning at exit under python -X dev
• pre-commit run --files on the changed files passes, mypy included