prediction fails when N_modegroups != N_modes

[nkern]

when N_modegroups != N_modes (if use_pca == True) or when N_modegroups != N_data (if use_pca == False), means that each for some (or all) regressors are fitting multiple target functions at once. This speeds up the training and prediction, but means that when we go to do E.predict(), we have to nest the resultant arrays carefully into the finalized weight (and by extension, recon) ndarray. This currently works for when each regressor gets its own target function, but breaks when the regressors try to take on more than one target function. i had this working before, so its just a matter of figuring out what broke it and how to fix it