Try extract waveform features with a spikeinterface backend

Spike interface offers a way to detect spikes and extract them from the raw data.
As of now, we are using heavy artillery with dartsorts and deep learning to denoise the waveforms, which slows down the computation.

Is it really worth it ? 
Let's try to extract waveform features using spike interface waveform extraction, and compare the training results on the EA dataset. 
