[FEATURE] Weight trimming to match a specific design effect

[talgalili]

Regarding balance/graviton (once we deal with the glmnet->sklearn transition).

How about we just move to using max_de to be based on weight trimming only (instead of shrinkage in the LASSO lambda stage)?
And if the results using weight trimming (for some max_de), are tolerable (similar to existing solution), then transitioning to it / maintaining it should be easier than the current CV solution.

TBD...