vibration_analyze: Add smooth_window to PSD fitting
Sometimes the broad resonant peak from the cantilever has sharper
unidentified peaks sitting on its shoulders. In the rare case that
these peaks are taller than the cantilever peak, they can throw off
the resonant frequency portion of the parameter-guessing heuristic.
By smoothing the PSD data somewhat, we can damp down the sharp peaks
and bring the cantilever peak back to the top.
Because the amount of smoothing you'll need depends on your noise
sources and the expected width of the cantilever peak, make the
smoothing configurable. A 10-sample-wide Hann filter (rescaled to
have unit weight) seems like a pretty conservative default.