Don't clear scalars in face tracking example (#4950)

This is what caused the deadlocks/crashes we were after (which I'm glad
it did -- those have been fixed in #4949).

Even though they are not a problem anymore, I'm not sure why they're
here to begin with? Clearing scalars does nothing unless you
specifically intend to use them with latest-at queries (i.e. not plots,
which isn't the case here), otherwise it's just inefficient with no
upside (which I'd rather not teach users). Am I missing something?
:thinking:

---

I've updated the code to only clear faces when they disappear.

I might come up with a follow up PR soon so that the timeseries view interprets those clears as plot breakpoints.

examples/python/face_tracking/main.py

@@ -237,13 +237,28 @@ def detect_and_log(self, image: npt.NDArray[np.uint8], frame_time_nano: int) ->
             if self._video_mode
             else self._detector.detect(image)
         )
-        rr.log("video/landmarker/faces", rr.Clear(recursive=True))
-        rr.log("reconstruction/faces", rr.Clear(recursive=True))
-        rr.log("blendshapes", rr.Clear(recursive=True))
+
+        def is_empty(i):  # type: ignore[no-untyped-def]
+            try:
+                next(i)
+                return False
+            except StopIteration:
+                return True
+
+        if is_empty(zip(detection_result.face_landmarks, detection_result.face_blendshapes)):
+            rr.log("video/landmarker/faces", rr.Clear(recursive=True))
+            rr.log("reconstruction/faces", rr.Clear(recursive=True))
+            rr.log("blendshapes", rr.Clear(recursive=True))
 
         for i, (landmark, blendshapes) in enumerate(
             zip(detection_result.face_landmarks, detection_result.face_blendshapes)
         ):
+            if len(landmark) == 0 or len(blendshapes) == 0:
+                rr.log(f"video/landmarker/faces/{i}/landmarks", rr.Clear(recursive=True))
+                rr.log(f"reconstruction/faces/{i}", rr.Clear(recursive=True))
+                rr.log(f"blendshapes/{i}", rr.Clear(recursive=True))
+                continue
+
             # MediaPipe's keypoints are normalized to [0, 1], so we need to scale them to get pixel coordinates.
             pts = [(math.floor(lm.x * width), math.floor(lm.y * height)) for lm in landmark]
             keypoint_ids = list(range(len(landmark)))