Classifier re-training on the fly for different waveform length

[abuzarmahmood]

Related to https://github.com/abuzarmahmood/blech_clust/issues/180

Currently the classifier pipeline only supports a single snippet length and a single sampling rate.
This will require re-sampling of the raw data to match the desired sampling rate and re-extraction of waveforms while maintaining the given labels.