Speed is set to 0 within half the filter window of nan-values

[MSchellenberger]

Within half the filtering window of a nan-value, the speed is set to 0. Since immobility is often used as a parameter in behavioral analysis, I think it would be better to have the speed set to np.nan instead.

Additionally, I found that if a dataset has multiple scorers that have the same bodypart, it also produces interesting intermediate result and throws an error. I know this rarely happens, but it should be easy to fix.

I visualized both issues in a jupyter notebook that I used for testing

I think I have a solution for both (see jupyter notebook) and would be happy to contribute!

[github-actions]

Welcome! Thank you for your contribution!