feat(cli): add depth verify/refine outputs and tests

- Retrieve depth + confidence measures from SVOReader when depth enabled
- Compute depth residual metrics and attach to output JSON
- Optionally write per-corner residual CSV via --report-csv
- Post-process refinement: optimize final pose and report pre/post metrics
- Add unit tests for depth verification and refinement modules
- Add basedpyright dev dependency for diagnostics

py_workspace/aruco/svo_sync.py

@@ -1,7 +1,7 @@
 import pyzed.sl as sl
 import numpy as np
 from dataclasses import dataclass
-from typing import List, Optional, Dict
+from typing import Any
 import os
 
 
@@ -13,18 +13,22 @@ class FrameData:
     timestamp_ns: int
     frame_index: int
     serial_number: int
-    depth_map: Optional[np.ndarray] = None
+    depth_map: np.ndarray | None = None
 
 
 class SVOReader:
     """Handles synchronized playback of multiple SVO files."""
 
+    svo_paths: list[str]
+    runtime_params: sl.RuntimeParameters
+    _depth_mode: sl.DEPTH_MODE
+
     def __init__(
-        self, svo_paths: List[str], depth_mode: sl.DEPTH_MODE = sl.DEPTH_MODE.NONE
+        self, svo_paths: list[str], depth_mode: sl.DEPTH_MODE = sl.DEPTH_MODE.NONE
     ):
         self.svo_paths = svo_paths
-        self.cameras: List[sl.Camera] = []
-        self.camera_info: List[Dict] = []
+        self.cameras: list[sl.Camera] = []
+        self.camera_info: list[dict[str, Any]] = []
         self.runtime_params = sl.RuntimeParameters()
         self._depth_mode = depth_mode
 
@@ -83,9 +87,9 @@ class SVOReader:
             else:
                 cam.set_svo_position(0)
 
-    def grab_all(self) -> List[Optional[FrameData]]:
+    def grab_all(self) -> list[FrameData | None]:
         """Grabs a frame from all cameras without strict synchronization."""
-        frames = []
+        frames: list[FrameData | None] = []
         for i, cam in enumerate(self.cameras):
             err = cam.grab(self.runtime_params)
             if err == sl.ERROR_CODE.SUCCESS:
@@ -110,7 +114,7 @@ class SVOReader:
                 frames.append(None)
         return frames
 
-    def grab_synced(self, tolerance_ms: int = 33) -> List[Optional[FrameData]]:
+    def grab_synced(self, tolerance_ms: int = 33) -> list[FrameData | None]:
         """
         Grabs frames from all cameras, attempting to keep them within tolerance_ms.
         If a camera falls behind, it skips frames.
@@ -168,14 +172,14 @@ class SVOReader:
     def enable_depth(self) -> bool:
         return self._depth_mode != sl.DEPTH_MODE.NONE
 
-    def _retrieve_depth(self, cam: sl.Camera) -> Optional[np.ndarray]:
+    def _retrieve_depth(self, cam: sl.Camera) -> np.ndarray | None:
         if not self.enable_depth:
             return None
         depth_mat = sl.Mat()
         cam.retrieve_measure(depth_mat, sl.MEASURE.DEPTH)
         return depth_mat.get_data().copy()
 
-    def get_depth_at(self, frame: FrameData, x: int, y: int) -> Optional[float]:
+    def get_depth_at(self, frame: FrameData, x: int, y: int) -> float | None:
         if frame.depth_map is None:
             return None
         h, w = frame.depth_map.shape[:2]
@@ -188,7 +192,7 @@ class SVOReader:
 
     def get_depth_window_median(
         self, frame: FrameData, x: int, y: int, size: int = 5
-    ) -> Optional[float]:
+    ) -> float | None:
         if frame.depth_map is None:
             return None
         if size % 2 == 0: