FilterReview: add auto fit for the first notch

[j-w9]

Adds an "Auto fit" button to the first notch that sets it up from the log: ESC telemetry first, then per motor throttle (then average throttle), with peak detection to verify the tracking and the noise floor setting the attenuation.

Method: the pre-filter spectrum of each FFT window in the analysis time is resampled against the tracked frequency and averaged. Noise that follows the motors adds up into peaks at multiples of the tracked frequency, and noise that doesn't is spread out.

• ESC is rejected if the largest peak is well away from a whole multiple of the ESC frequency, with a message to check the motor poles.
• Harmonics whose peak is at least 6 dB above the local noise floor are enabled. ATT is how far the largest one stands out, limited to 10–50 dB.
• Rules of thumb: ESC FREQ is 80% of the lowest motor frequency seen. For throttle, REF is the mean throttle and FREQ is fitted from the peaks, and FM_RAT is lowered to keep tracking below REF. BW is FREQ/2. OPTS is set to multi-source with a single notch per harmonic.

These are all in one block at the top of AutoFit.js and are the main thing I'd like your view on.

Tested on six batch-logged copter logs, with estimated noise above 30 Hz:

• On the four flights where it picked ESC, the fit matched or beat the existing hand-set notch.
• On a ground run, it reports the ESC frequency mismatch and that throttle is zero.
• Synthetic spectra with known peaks recover the harmonics and attenuation, and a 14/12 pole error is caught.

Known limitation: a fixed resonance that sits near a motor harmonic in hover can be picked up as a harmonic.