SpherePad for equi2cube + cache for create_equi_grid and create_normalized_grid

[antoine-scenario]

2 new contributions:

[Feature] - SpherePad feature for equi2cube

Implements Bifuse SpherePad for equi2cube transform.
Usage:

equilib.equi2cube(equi, ..., pad=pad)

with pad a float between 0 (no overlap) and 1 (overlap with the entire adjacent faces).

[Optimization] - Cache for create_equi_grid and create_normalized_grid

When calling cube2equi or equi2equi on loop with the same input and ouptut dimensions, grid computation is done once for all, which saves a lot of time.

[haruishi43]

@antoine-scenario looks good!
Just one thing. It seems that functools.cache was introduced prior to python 3.9, so the library fails for 3.8.
Do you know/see if functools.cache has significant improvements over functools.lru_cache?
If you think we should move on to cache, I might want to create a wrapper around cache and lru_cache inside equilib.utils so we can automatically detect the python version and use the best caching function.