Encoder nonlinearity compensation and calibration robustness

[datacrystals]

This issue tracks work to fix encoder calibration false-triggers, improve measurement quality under rotor slip, and lay groundwork for encoder nonlinearity (INL) compensation.

Fixes already on branch feature/encoder-distortion-compensation

• BreakawayFinder zero-mod phantom-motion guard — rejects encoder movement reported while modulation is near zero, so floating/disconnected sin/cos inputs can no longer false-trigger breakaway at mod=0.000.
• cal --no-save flag — lets users run calibration routines without writing results to FRAM/RTE KV store.
• FRAM failure messaging — AutoCalibrationCoordinator::finish() now reports exactly which persistence path failed instead of printing a success message after an FRAM error.
• Locked-region encoder cycles/rev — EncoderOffsetCalibrator now measures cycles/rev only over the verified-locked portion of the rotation, eliminating the corrupted 0.789 value seen when the rotor slips during the initial segment.
• Slip detection / longer lock timeout — warmup and offset-rotate logs include lock ratio; calibration waits up to 450° of field rotation for the rotor to lock before failing.

Next: encoder nonlinearity / distortion compensation

The learned encoder bounds show a ~5,100-count DC offset mismatch between sin and cos channels, indicating a 1st-harmonic angle error in the current decode. The next step is to build a per-rev correction layer:

1. Ellipse-fit / harmonic correction for amplitude imbalance, DC offset, and quadrature error.
2. Optional LUT or Fourier-based per-angle compensation.
3. Validate against enc_trace and FOC torque ripple.

Branch: https://github.com/OpenVVVF/RTE/tree/feature/encoder-distortion-compensation

[datacrystals]

high speed causes instability. fixing

[datacrystals]

Status update — d44975d, on main via ec4907b / 1704d0e.

Delivered:

• BreakawayFinder phantom-motion guard: encoder "movement" at mod≈0 now fails the cal early instead of declaring breakaway=0.000 (root cause of the doomed offset runs)
• --no-save flag, honest FRAM failure messaging, locked-region-gated cycles/rev measurement (0.789 → ~1.00), slip/lock diagnostics in warmup
• FRAM motor-config save fixed (MotorConfigStore::init() restored)
• Ellipse fit (Layer 1) implemented and persisted under Motor.Encoder.SinCos.Fit.*
• ENCLIN measurement harness: step-and-settle INL sweep with FWD+REV dual-pass lag cancellation; first harmonic tables produced

Not done / do not trust yet:

• Fit phi is not repeatable (1.25°–3.7° across consecutive runs); a bad saved phi measurably degrades commutation. Fit stays cleared until capture is lock-gated and phi repeats within ±0.2°.
• Layer 2 INL correction: ENCLIN measures the residual but the correction table itself is not implemented; single-direction residuals are dominated by torque-lag/cogging (dual-pass cancellation implemented but unvalidated).
• See Make calibration fully voltage-independent #37 for voltage-independence and remaining items.

Leaving open until the compensation path is actually validated on hardware.