Should `get_traces` with `return_scaled=True` when the base dtype is float.

[h-mayorquin]

From the discussion inhttps://github.com//pull/2864

We have the following lines:

https://github.com/h-mayorquin/spikeinterface/blob/bc0cbd9f74f30cfed9376f984091cd57f3634169/src/spikeinterface/core/baserecording.py#L369-L373

But if the original dtype is float64 this will downcast. Maybe we should change the code a little bit.

[JoeZiminski]

Agree it makes sense if the dtype of the recording is already float64 to have get_traces return float64 when get_traces is used. So overall if I understand SI behaviour correctly the (new) system would be:

Original	Unscaled	Scaled
int	int	float32
float32	float32	float32
float64	float64	float64

[h-mayorquin]

Yes, int should be cast to float.

I think that float should not be cast at all. Maybe float16 if that appears for some reason? Does it appear?

[JoeZiminski]

Good point, yes float16 seems to be supported and would agree it makes sense that the dtype for any float should be preserved between return_scaled and return_unsclaed.