tracker: release manual control when setting mode

[robertlong13]

When I was testing ArduPilot/ardupilot#30202, it took me a while to figure out how to release manual control. Once you do tracker position, you cannot get the tracker to do anything else until you call tracker position 32767 32767.

I think it's probably appropriate to release manual control any time the user changes mode, so that's what I've done here.

[tridge]

Automated review note — AI-generated (Claude), validated against the live diff. Please sanity-check before acting.
Full report: https://uav.tridgell.net/DevCallReviews/2026_09_01_AIReview/devcall_pr_reviews.html#prMAVProxy-1575

Reviewed at head b8dd2d249c. COMMENT — the premise is right and it still applies cleanly to master. Nothing in AntennaTracker clears manual_control_yaw/_pitch on a mode change, 32767 is exactly the firmware's 0x7FFF sentinel, and positions[0]→yaw / positions[1]→pitch matches the firmware's field use.

One ordering issue. cmd_tracker_position assigns self._manual_control at line 103 and only then calls manual_control_send() at 104. If the send raises on the release path, the flag has already been cleared, the exception propagates out of the nested call so connection.set_mode() on the next line never runs, and a retry won't re-attempt the release because the flag now says there's nothing to release — the mode change is lost and the state is wrong. The take path has the mirror problem but is harmless (a spurious release is a no-op). Assigning the flag after a successful send, or a try/finally, closes both.

For transparency: my two review passes disagreed on this one — one called it a blocking bug, the other a minor note. Both described the same mechanics, so I've landed between them: a real ordering defect with a trivial fix, but it needs the send to actually raise, and this is an antenna tracker rather than a vehicle.

Two smaller notes. The generic mode command bypasses the release entirely — with the tracker selected as the current link, mode AUTO goes through mavproxy_mode.py's cmd_mode (:41, :74) straight to master.set_mode(), leaving the axes latched. Out of scope for a 6-line fix, but worth a follow-up. And find_connection() is re-resolved inside the nested call and prefers self.connection, so the release can go down a different link than the one that took manual control.