UI refinement

Revert to window-mean gravity

Fix filtfilt padlen for 0.3Hz gravity filter, default 9-sample pad inadequate, use max safe pad

Proper UCI pipeline: filter full signal before windowing, revert norm_params to UCI body_acc

Fix gravity separation: use window mean instead of 0.3Hz LP (filter needs 167 samples to settle, window is only 128); recompute norm_params to match

Fix accelerometer unit conversion: divide Phyphox m/s² by 9.80665 to match UCI g units

wire up Phyphox pipeline: full 561 feature extraction from raw CSV