account for multiplicity in spot prediction/res-dep BW

[DHekstra]

since we have miller indices and spacegroup by the time we predict spots, we know the multiplicities of each spot. perhaps there is merit in adjusting the threshold accordingly. (granting that will require some thought to determine what "accordingly" means)

[kmdalton]

This should be straightforward to implement. I'm not sure if it'll make much difference, but it's an interesting idea.

[DHekstra]

would it make sense to replace the KDE with logistic regression? with a set of strong reflections and a set of not strong reflections as the training data and wavelength, rlp, multiplicity, and possibly the image number as predictive variables?