Fix dtypes handled in the labels of the decoders

[CeliaBenquet]

int and float subtypes accepted to be in the labels of the decoders were hard coded, which results in errors for int8 for instance.

Now, it supports all subtypes of Integer and floating in numpy arrays and pytorch tensors.

• adapted the corresponding tests.

Fix https://github.com/AdaptiveMotorControlLab/CEBRA-dev/pull/616

[stes]

Seems like the documentation status still fails due to

I'll disable the comment function for now so we can merge this and other upcoming PRs, the check is still executed.