feat: add --tracemalloc flag for memory profiling

Starts tracemalloc before the event loop and dumps the top 25
allocations on shutdown. Accepts optional nframes depth (default 10).
Can be combined with --cprofile.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

src/derp/cli.py

@@ -40,6 +40,14 @@ def build_parser() -> argparse.ArgumentParser:
         const="derp.prof",
         help="enable cProfile; dump stats to PATH [derp.prof]",
     )
+    p.add_argument(
+        "--tracemalloc",
+        metavar="NFRAMES",
+        nargs="?",
+        const=10,
+        type=int,
+        help="enable tracemalloc; capture NFRAMES deep [10]",
+    )
     p.add_argument(
         "-V", "--version",
         action="version",
@@ -70,6 +78,23 @@ def _shutdown(bot: Bot) -> None:
     asyncio.get_running_loop().create_task(bot.conn.close())
 
 
+def _dump_tracemalloc(log: logging.Logger, limit: int = 25) -> None:
+    """Log top memory allocations from tracemalloc snapshot."""
+    import tracemalloc
+
+    snapshot = tracemalloc.take_snapshot()
+    snapshot = snapshot.filter_traces([
+        tracemalloc.Filter(False, "<frozen importlib._bootstrap>"),
+        tracemalloc.Filter(False, "<frozen importlib._bootstrap_external>"),
+        tracemalloc.Filter(False, "<unknown>"),
+    ])
+    stats = snapshot.statistics("traceback")
+    total = sum(s.size for s in stats)
+    log.info("tracemalloc top %d (total tracked: %.1f KiB)", limit, total / 1024)
+    for i, stat in enumerate(stats[:limit], 1):
+        log.info("#%d  %.1f KiB  %s", i, stat.size / 1024, stat.traceback.format()[0])
+
+
 def main(argv: list[str] | None = None) -> int:
     """Main entry point."""
     parser = build_parser()
@@ -93,6 +118,12 @@ def main(argv: list[str] | None = None) -> int:
     bot = Bot(config, registry)
     bot.load_plugins()
 
+    if args.tracemalloc:
+        import tracemalloc
+
+        tracemalloc.start(args.tracemalloc)
+        log.info("tracemalloc enabled, nframes=%d", args.tracemalloc)
+
     if args.cprofile:
         import cProfile
 
@@ -102,6 +133,9 @@ def main(argv: list[str] | None = None) -> int:
     else:
         _run(bot)
 
+    if args.tracemalloc:
+        _dump_tracemalloc(log)
+
     return 0