Add limits that cleverly enforce the size of datasets that can be used

[samjmolyneux]

Both hyperparameter optimisation and shap values stop working properly when data gets too large.

Need to figure out why for both and find a way to do clever validation.

[samjmolyneux]

Surely it doesnt make sense that we are having these problems with the Shap plotter, given that it supposedly can deal with massive datasets.