Fix event loop blocking from perf endpoint and profile detection

- Change /api/v1/system/performance from async def to def so FastAPI
  runs the blocking psutil + NVML GPU queries in a thread pool instead
  of freezing the event loop (polled every 2s by dashboard)
- Batch profile engine's 3 separate run_in_executor detection calls
  into a single _detect_all_sync() call, reducing event loop wake-ups

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

server/src/wled_controller/api/routes/system.py

@@ -140,8 +140,13 @@ async def get_running_processes(_: AuthRequired):
     response_model=PerformanceResponse,
     tags=["Config"],
 )
-async def get_system_performance(_: AuthRequired):
-    """Get current system performance metrics (CPU, RAM, GPU)."""
+def get_system_performance(_: AuthRequired):
+    """Get current system performance metrics (CPU, RAM, GPU).
+
+    Uses sync ``def`` so FastAPI runs it in a thread pool — the psutil
+    and NVML calls are blocking and would stall the event loop if run
+    in an ``async def`` handler.
+    """
     mem = psutil.virtual_memory()
 
     gpu = None

server/src/wled_controller/core/profiles/profile_engine.py

@@ -68,6 +68,23 @@ class ProfileEngine:
         async with self._eval_lock:
             await self._evaluate_all_locked()
 
+    def _detect_all_sync(
+        self, needs_running: bool, needs_topmost: bool, needs_fullscreen: bool,
+    ) -> tuple:
+        """Run all platform detection in a single thread call.
+
+        Batching the three detection calls into one executor submission reduces
+        event-loop wake-ups from 3 to 1, minimising asyncio.sleep() jitter in
+        latency-sensitive processing loops.
+        """
+        running_procs = self._detector._get_running_processes_sync() if needs_running else set()
+        if needs_topmost:
+            topmost_proc, topmost_fullscreen = self._detector._get_topmost_process_sync()
+        else:
+            topmost_proc, topmost_fullscreen = None, False
+        fullscreen_procs = self._detector._get_fullscreen_processes_sync() if needs_fullscreen else set()
+        return running_procs, topmost_proc, topmost_fullscreen, fullscreen_procs
+
     async def _evaluate_all_locked(self) -> None:
         profiles = self._store.get_all_profiles()
         if not profiles:
@@ -84,26 +101,18 @@ class ProfileEngine:
                     mt = getattr(c, "match_type", "running")
                     match_types_used.add(mt)
 
-        # WMI process enumeration (~3s) — only needed for "running" match type
         needs_running = "running" in match_types_used
-        running_procs = (
-            await self._detector.get_running_processes()
-            if needs_running else set()
-        )
-
-        # Foreground window check (<1ms) — needed for "topmost" and "topmost_fullscreen"
         needs_topmost = bool(match_types_used & {"topmost", "topmost_fullscreen"})
-        if needs_topmost:
-            topmost_proc, topmost_fullscreen = await self._detector.get_topmost_process()
-        else:
-            topmost_proc = None
-            topmost_fullscreen = False
-
-        # Fullscreen window enumeration (<1ms) — only needed for "fullscreen"
         needs_fullscreen = "fullscreen" in match_types_used
-        fullscreen_procs = (
-            await self._detector.get_fullscreen_processes()
-            if needs_fullscreen else set()
+
+        # Single executor call for all platform detection (avoids 3 separate
+        # event-loop roundtrips that can jitter processing-loop timing)
+        loop = asyncio.get_event_loop()
+        running_procs, topmost_proc, topmost_fullscreen, fullscreen_procs = (
+            await loop.run_in_executor(
+                None, self._detect_all_sync,
+                needs_running, needs_topmost, needs_fullscreen,
+            )
         )
 
         active_profile_ids = set()