[Feat] working sinkhorn version and gradient checkpointing in batch dimension

[SCMusson]

A functioning sinkhorn_trainer class Gradient checkpointing module in molearn/utils.py

Two example scripts that hopefully demonstrate both of the above techniques