feat!: reorganize detection and tracking pipeline

Refactor the package into common, schema, detection, and tracking namespaces and move dataset-specific ActualTest utilities into tests/support.

Add a pluggable detection stack with typed protocols, pydantic-settings config, loguru-based runner logging, cvmmap and headless video sources, NATS and parquet sinks, and a structured coco-wholebody133 payload path.

Teach tracking replay loading to consume parquet detection directories directly, preserve empty frames, and keep the video-to-parquet-to-tracking workflow usable for offline E2E runs.

Vendor the local mmcv and xtcocotools wheels under Git LFS, update uv sources/lock state, and refresh the mmcv build so mmcv.ops loads successfully with the current torch+cu130 environment.

tests/test_kinematics_seed.py

@@ -1,7 +1,7 @@
 import numpy as np
 
-from pose_tracking_exp.joints import BODY20_INDEX_BY_NAME
-from pose_tracking_exp.kinematics import seed_state_from_pose3d
+from pose_tracking_exp.common.joints import BODY20_INDEX_BY_NAME
+from pose_tracking_exp.tracking.kinematics import seed_state_from_pose3d
 
 
 def _sample_pose3d() -> np.ndarray:
@@ -38,7 +38,7 @@ def test_seed_state_from_pose3d_does_not_call_least_squares(monkeypatch) -> None
     def fail_least_squares(*args: object, **kwargs: object) -> object:
         raise AssertionError("seed_state_from_pose3d should not call scipy.optimize.least_squares")
 
-    monkeypatch.setattr("pose_tracking_exp.kinematics.least_squares", fail_least_squares)
+    monkeypatch.setattr("pose_tracking_exp.tracking.kinematics.least_squares", fail_least_squares)
     state = seed_state_from_pose3d(_sample_pose3d())
 
     assert state.parameters.shape == (31,)