feat: auto-checkpoint on SIGTERM/SIGINT with DDP support

- Register SIGTERM/SIGINT handlers in training loop, set stop flag on signal
- Check stop_requested at each epoch/batch boundary, break and call on_error to save checkpoint
- LocalStrategy parent forwards signal to child processes via terminate(), waits up to 600s for graceful exit
- TrainContext gains threading.Event-based stop_requested/request_stop
- Tests verify SIGTERM/SIGINT trigger checkpoint save with exit code 0, works on both CPU and GPU

astrai/parallel/signal_handler.py

@@ -0,0 +1,44 @@
+import logging
+import os
+import signal
+import threading
+
+logger = logging.getLogger(__name__)
+
+_early_stop = threading.Event()
+_active_context = None
+
+
+def _early_handler(signum: int, frame):
+    sig = signal.Signals(signum)
+    logger.warning(
+        "Received %s (pid=%d), requesting graceful training stop...",
+        sig.name,
+        os.getpid(),
+    )
+    _early_stop.set()
+    if _active_context is not None:
+        _active_context.request_stop()
+
+
+def install_early_signal_handlers():
+    for sig in (signal.SIGTERM, signal.SIGINT):
+        signal.signal(sig, _early_handler)
+
+
+def register_signal_handlers(context):
+    global _active_context
+    _active_context = context
+    for sig in (signal.SIGTERM, signal.SIGINT):
+        signal.signal(sig, _early_handler)
+    if _early_stop.is_set():
+        context.request_stop()
+        logger.warning("Signal was received during initialization, stopping...")
+
+
+def unregister_signal_handlers():
+    global _active_context
+    _active_context = None
+    _early_stop.clear()
+    signal.signal(signal.SIGTERM, signal.SIG_DFL)
+    signal.signal(signal.SIGINT, signal.SIG_DFL)