Should the rescaling use Q2/Q98 instead of min/max ?

[fpavogt]

@loforest suggested that when looking for rescaling bounds in ampycloud (e.g. for the slicing), it might be beneficial to rely and quantiles rather than min/max, in order to be more robust against outliers.

I'm leaving this here as an idea for future improvements to the code.