Add regression tests for LiDAR hits, CV/FastBEV perception, and control dual-write

[cursor]

Summary

Adds focused regression coverage for high-risk paths that remained untested on dev, while avoiding duplication of open drafts #17/#20/#22–#30.

Risky behavior now covered

• BasicSim LiDAR raycasting: empty scenes; forward hits on boundary walls and agent BB near faces; misses beyond c46_lidar_range; nearest-hit selection when wall and agent both intersect a ray
• ConstantVelocityPrediction.predict: empty-agent clear; horizon/heading extrapolation keyed by agent_id; None perception model raises
• FastBEVLidarDetection.detect: gap > mu splits clusters; diagonal outside [delta_min, delta_max] dropped; edge-on/collinear clusters get default L/W boxes pushed away from ego; 3D z-band filter clears/accepts points
• VisualizerApp.apply_world_control: after world/stack ego split, Control Step/Steer advances the plant then copies into pm.ego_vehicle in place

Test files added/updated

• test/c40_execution/test_c46_basic_sim_lidar.py (new)
• test/c10_perception/test_c15_constant_velocity_prediction.py (new)
• test/c10_perception/test_c15_fast_bev_detection.py (new)
• test/plugins/test_p60_visualizer_tk/test_p61_apply_global_plan.py

Why these tests materially reduce regression risk

BasicSim LiDAR feeds FTG, FastBEV, and lidar localization in the default sim bridge—silent ray math bugs corrupt every LiDAR consumer. Constant-velocity prediction is the default predictor in bundled profiles; wrong horizon/heading changes lattice and velocity collision sweeps. FastBEV is the active detector in perception.yaml with complex segmentation/MBR edge cases. apply_world_control is the merged fix path for Control Step/Steer desync after the world/stack ego split.

Validation

pytest \
  test/c40_execution/test_c46_basic_sim_lidar.py \
  test/c10_perception/test_c15_constant_velocity_prediction.py \
  test/c10_perception/test_c15_fast_bev_detection.py \
  test/plugins/test_p60_visualizer_tk/test_p61_apply_global_plan.py -v

All 17 tests passed. No production code changes.