Implement handling of non-three-point water models

[davidlmobley]

Longer-term, we probably want to implement the ability to handle non-three-point water models. This will take some substantial work, since amber.build_mixture_prmtop and packmol.pack_box work typically from .mol2 files (i.e. for water) and for four or five (or other) point water models we won't have a .mol2 file to work from, and there will be other architectural changes as well.

142 addresses issue #134 for three-point water models, but more general water models will require more rethinking.

I would love input on the best way to do this. Changes will be needed in at least two places:

• amber.build_mixture_prmtop; probably this can be handled there by requiring the user to specify which component is water and requiring the mol2_filenames and frcmod_filenames to be shorter. But there may still be issues relating to residue/atom naming
• packmol.pack_box: We'll probably need to modify this to know the target water model (for non-three-point models) and have sample water geometry files with the expected atom/site names (expected by tleap) and number of sites, and work from these rather than .mol2 files when an N != 3 site model is desired.

[davidlmobley]

A likely solution is to use the approach mentioned by @jchodera in #142 to help with this. Specifically:

Note that simtk.openmm.app.Modeller offers a method addExtraParticles to add any missing extra sites for multipoint waters defined with OpenMM ffxml files. I'm not sure if it is feasible to utilize this in the pipeline or not.