Fix Sync ZOH after Control off and Frenet SD orientation

[cursor]

Summary

Two high-severity correctness bugs found in recent 0.5.3-era control/Frenet paths:

1. SyncExecuter keeps actuating after Control is unchecked — _simulate_step continued applying the held ZOH command when call_control=False, so the plant kept integrating the last accel/steer. Async already exits its control worker when Control is off; Sync did not.
2. Frenet teleport/spawn orientation uses tracker current_wp, not path heading at s — convert_sd_orientation_to_xy_orientation added the heading of the currently tracked segment. Teleporting/spawning on a later segment (e.g. after a corner) pointed the ego/NPC the wrong way. The Frenet-view orient drag also passed the mouse-tip (s,d) instead of the click pose.

Root cause

• 0.5.3 split control compute from plant integrate (ZOH). Simulate always applied _last_cmd whenever actuation was allowed, ignoring the Control stage gate.
• Orientation conversion reused tracker waypoint indices instead of the same arc-length segment bracketing as convert_sd_to_xy.

Fix

• Clear _last_cmd when call_control=False (paced ZOH with Control still enabled is unchanged).
• Bracket path heading by s in convert_sd_orientation_to_xy_orientation.
• Frenet orient drag converts at teleport_s/teleport_d.

Validation

• test_call_control_false_drops_held_zoh_command — Control off → no further control_ego_state, _last_cmd cleared; existing paced-ZOH test still passes.
• test_sd_orientation_uses_heading_at_s_not_tracker_wp — vertical-leg query with tracker at start → θ ≈ π/2; also covers final-wp current==next.