kde_histogramdd, test need to be made more generic and have hacks removed

[jllanfranchi]

• requires specific binning dimension names
• reordering dimensions for PID is performed in ad hoc, arcane fashion
• energy units are forced to be GeV or it fails
• must be 'pid' dimension and two other dimensions, not clear if those have to be 'coszen' and 'energy' or just two other dimensions, but 'energy' does have to have units of GeV if it is present, apparently
• calls "dimensions" bins (rather than "dimensions") in code just to make it confusing
• unit test is very poorly constructed to show what it's doing and why (silly hacks to get random numbers in a specific range? ; no repeatability since no seed values set; etc.)
• unit test fails, but rather than just hacking it further, it should be cleaned up

[philippeller]

Still not a beauty, but errors in tests removed and superficially cleaned up in https://github.com/jllanfranchi/pisa/pull/289

[LeanderFischer]

Possibly realted to #90 and #108. But there is a lot of stuff that relies on this specific binning.. not great 😔