profiler: add per-thread cProfile support

threading.setprofile installs a bootstrap that creates a per-thread
cProfile.Profile.  Stats from all threads are merged on periodic
dumps and at exit, capturing worker-thread hotspots (audio send
loop, pitch shifting).

src/tuimble/__main__.py

@@ -31,10 +31,17 @@ def main():
 
 
 def _run_profiled(app, dest):
-    """Run the app under cProfile with periodic 30s dumps."""
+    """Run the app under cProfile with periodic 30s dumps.
+
+    Profiles both the main thread and all worker threads (including
+    the audio send loop where PitchShifter.process runs).  Each
+    thread gets its own cProfile.Profile; stats are merged on dump.
+    """
     import cProfile
+    import pstats
+    import threading
     from pathlib import Path
-    from threading import Event, Thread
+    from threading import Event, Lock, Thread
 
     if dest is None:
         from tuimble.config import CONFIG_DIR
@@ -47,10 +54,34 @@ def _run_profiled(app, dest):
 
     prof = cProfile.Profile()
     stop = Event()
+    thread_profiles: list[cProfile.Profile] = []
+    lock = Lock()
+
+    def _thread_bootstrap(frame, event, arg):
+        """Installed via threading.setprofile; creates a per-thread profiler.
+
+        Called once per new thread as the profile function.  Immediately
+        creates and enables a cProfile.Profile whose C-level hook
+        replaces this Python-level one for the remainder of the thread.
+        """
+        tp = cProfile.Profile()
+        with lock:
+            thread_profiles.append(tp)
+        tp.enable()
+
+    threading.setprofile(_thread_bootstrap)
+
+    def _dump_all():
+        """Merge main + thread profiles and write to disk."""
+        stats = pstats.Stats(prof)
+        with lock:
+            for tp in thread_profiles:
+                stats.add(tp)
+        stats.dump_stats(str(dest))
 
     def _periodic_dump():
         while not stop.wait(30):
-            prof.dump_stats(str(dest))
+            _dump_all()
 
     dumper = Thread(target=_periodic_dump, daemon=True)
     dumper.start()
@@ -60,8 +91,12 @@ def _run_profiled(app, dest):
         app.run()
     finally:
         prof.disable()
+        threading.setprofile(None)
+        with lock:
+            for tp in thread_profiles:
+                tp.disable()
         stop.set()
-        prof.dump_stats(str(dest))
+        _dump_all()
 
 
 if __name__ == "__main__":