Batch size memory check for explainer algorithms

[arnaudvl]

Different explainer algorithms (Anchors, numerical gradient eval of Counterfactuals) require batch model predictions. There is currently no check in place whether these would fit in memory, and if they don't, how to distribute them.