Improve ROI LOS rates and align camera telemetry timing

[tridge]

Vehicle and gimbal telemetry can describe different instants, producing yaw-rate ripple during ROI tracking. Finite differences of rounded positions and gimbal angles also introduce noise into control feed-forward and SITL video prediction.

Map vehicle boot timestamps into camera monotonic time using the AP_RTC minimum-delay estimator, and interpolate vehicle yaw/rate at the gimbal feedback timestamp. Derive stationary ROI line-of-sight rates analytically from GLOBAL_POSITION_INT position and NED velocity, and carry measured gimbal yaw rate through video metadata for prediction. Predictions stop after 250 ms of stale data; ATTITUDE fallback becomes eligible at that limit. Legacy metadata without measured rates remains supported.

Includes regression coverage for timestamp wrapping, restarts, reordered/buffered samples, fallback, singular geometry and quantised gimbal feedback, plus camera and SITL documentation updates.

Validation:

• MT11 SITL build and targeted telemetry-time, targeting, MAVLink-time and video-metadata tests passed.
• Normal and jittered ROI angle tracking, jittered ROI rate tracking, MAVLink parameters and MT11 angle-hold integration checks passed.
• Terrain video geometry, prediction, loading and encoding tests passed. The synthetic quantisation test reduced frame-to-frame yaw prediction jitter about fourfold with measured rates.
• The broader video telemetry integration test has a zoom/FOV assertion failure reproduced on the baseline. Hardware validation and confirmation of the final change in the live flight remain outstanding.

The clock estimator cannot remove fixed one-way latency; MT11 feedback is still timestamped on receipt. Real motor-response ripple can remain.

[tridge]

replaced by #31