Fix closed-loop path_s corruption and Frenet corner false CTE

[cursor]

Bug and impact

1. Closed race tracks corrupt path_s — When first==last (bundled Yas Marina race line, closed RaceMap centerlines), KD-tree Frenet re-projection of the reference snapped the final waypoint to index 0, yielding non-monotonic path_s with path_s[-1] == 0 (lattice already worked around this via path_s[-2]).
2. Pure Pursuit aimed at start/finish for the whole lap — find_path_lookahead clamped with min(s+Ld, path_s[-1]). With path_s[-1]==0, every mid-lap lookahead jumped ~1.2 km to the origin on Yas Marina.
3. False CTE after corners — XY→SD always projected onto the segment ending at the nearest waypoint, so an on-path point after a 90° turn could report d≈segment_length instead of d≈0, yanking Stanley/PID/PP.

Root cause

• Reference path_s was computed by re-projecting waypoints through convert_xy_path_to_sd_path instead of using cumulative arc-length.
• Frenet segment selection ignored the outgoing segment around the nearest vertex; SD→XY used nearest-waypoint rather than arc-length bracketing.

Fix

• Initialize path_s / path_d from __cumulative_distances.
• XY→SD scores both adjacent segments (clamped distance); SD→XY brackets by searchsorted on arc-length; 1-point paths handled safely.
• Pure Pursuit clamps with max(path_s).

Validation

PYTHONPATH=/workspace python3 -m pytest \
  test/c50_common/test_c54_closed_loop_path_s.py \
  test/c30_control/test_c35_pure_pursuit.py \
  test/c50_common/test_c54_trajectory_horizon_slice.py \
  test/c50_common/test_c54_trajectory_waypoint_update.py \
  test/c50_common/test_c54_trajectory_tracker_concatenate.py \
  test/c30_control/test_c33_pid.py -v

42 passed.

Not re-reporting already-tracked PR #18 (1-point Frenet / Control Step) or merged PR #16.

  

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.