fix: fixed calculation of differences

[manuba95]

This fixes the unwanted behavior of the numpy gradient and diff functions to replace the first/last value of a NaN-free sequence with a NaNs. The differentiated data now has values where the original data has values and vice versa. This is achieved by the same approach we used to get the filterable sequences for the low pass filter.
Additionally, a minor bug in the call of the numpy diff function was fixed.

[draabe]

Hi, thanks for pointing this out.

As we are having similar considerations in the filter functions or ApEn calculation, I think this requires a bigger refactor for
1.) generalizing the functionality for sequence-finding (refactoring the private _get_filterable_and_short_sequences function) and
2.) column- and sequence wise application of functions.

So I'm making this a draft for now and put this on the priority list for the next major update!