More SKL dependence

[bgpeyton]

Previously issue #35 from lml.

While the home-coded ml algorithms have mostly been phased out of default use, there are still a few defaults remaining that use mlqm functions where skl would suffice. A good example is that, generally, at the end of a kernel-ridge regression run the krr.predict function is used, and passed in a Dataset object along with the representations of the prediction set.

E = mlqm.krr.predict(ds,val_reps)

However, this means that mlqm is used to 1) generate the kernel and 2) solve the set of linear equations for the target function. While this function has been tested to be equivalent to skl, it is far more limiting in that different kernels are not supported in mlqm and additional options that skl rolls out will not be supported unless specifically baked into mlqm. So instead, this function should simply wrap skl with optional kwargs rather than doing its job, so as to future-proof the code and automatically gain any additional functionality of skl.