Improvement: restructure caching logic and handling

[pzeiger]

Right now the caching logic is quite convoluted. I would suggest to simplify it in some way. Also MultsliceCalculator does only cache exit waves as far as I can see. It does not cache the potential or slices, which are named as cache_levels.

[tpchuckles]

• exitwaves are cached by MultisliceCalculator.run directly (this is the final wave).
• slices are cached by expanding the variable "wavefunction_data" to store the layer-by-layer exit wave (which is a bit hard to follow in the code).
• potentials are cached by Potentials, inferred based on the passing of the "cache_dir" argument

I agree it's definitely a bit of a mess, as we have duct-taped things on as we went.