Robustness with respect to \alpha / GW vs W about moscot

[Marius1311]

Basically, moscot converges to a completely different GW regime than POT does.

[Marius1311]

Further checks for me:

• check whether for alpha \sim 0, we approach entropic. reg. W
• check whether for alpha \sim 1, we approach entropic reg. GW

[Marius1311]

Update: I included the above two in my original issue post.

[Marius1311]

Update: I added actual novosparc to the comparison above.

[Marius1311]

Notes

• running in log space?
• increasing numerical accuracy? i.e. using float64
• adapting epsilon (using OTTs heuristic?)

[Marius1311]

This is done, we gained some intuition.