[Feature Request] Experimental MC autotune for difficult-to-tune vehicles

Problem

Standard MC Autotune may fail to complete on some aircraft: its experiment timeout is fixed, and coefficient covariance does not directly verify the proposed controller change.

Preferred solution

Explore an opt-in experimental method alongside standard Autotune for these cases. Candidate gains are checked against independent response measurements, replacing the fixed covariance convergence threshold; measurement timing is configurable for slow, high-inertia dynamics, including those of larger vehicles.

PR #28706 now preserves standard Autotune by default and selects the experimental implementation at build time with CONFIG_MC_AUTOTUNE_EXPERIMENTAL. The dedicated px4_sitl_autotune-experimental configuration enables it for SITL; there is no runtime method selector.

Alternatives

Investigate and fix specific failures in standard Autotune independently. The experimental method would remain optional: it takes longer, uses more resources and currently requires Position-mode hover.

Results

X500 and an inertia ×4 model successfully completed SITL tuning, retained their gains after full PX4 restarts and completed subsequent hovers; the original method timed out in these scenarios. The synthetic 1000 kg model completed measurement but was rejected by existing gain bounds, so successful tuning of that model is not claimed.

Detailed comparison and reproduction steps. Lifecycle fixes are tracked separately in #28699 / #28700.