Better way to deal with scaling

The normalized wavelet transform doesn't work well if the raw values are too small. Right now, things are rescaled by a constant factor in [store_resample](https://github.com/BouchardLab/process_nwb/blob/94e1f4ac4ed0f518da1c59c09d0effe1746c9506/process_nwb/resample.py#L152). This should probably be done somewhere else and maybe with an adaptive value?