Fix limits when numbers are all NaN and we have log scale

[nvaytet]

This PR fixes the following:

import plopp as pp

a = pp.data.data2d()
a.values[...] = float("nan")

pp.plot(a, norm='log')

It creates a blank figure without raising an error.

[jl-wynen]

Are lims[0] and lims[1] guaranteed to have the same unit? The operations below would fail if that is not the case. (This is not new)

[nvaytet]

As long as fix_empty_range is called on the output of find_limits, this is guaranteed, because find_limits takes in only a single Variable, which has the same unit for all the values it holds.

If fix_empty_range suddenly gets used elsewhere, then you could have that error. I can add a check, I guess it's not too expensive (even though everything adds up when you do many small checks on every plot update).