Conformer generation and validation cleanups

[j-wags]

Questions

• [ ] What RMS cutoff was used in the original paper? (I've just fired from the hip and set it to 2A here)
• [ ] (Kinda for my previous PR) How do I make click accept multiple arguments for a single keyword?
  • nargs=-1
• [ ] How would we want to structure a general "dataset" class for this project? Pandas dataframe?
  • Don't standardize yet -- Dict of dicts is fine
• [ ] Should we NOT output smiles from this step? (since any successful mol must have a conf at this point)

Remaining to-dos

• [x] Wire up CLI
• [ ] RDKitToolkitWrapper is doing some nasty things with bond stereo -- This seems like a major toolkit bug. I worked on it a bit but it'll need a kinda deep fix [1]
• [x] ~Add toolkit from_file functionality to skip/separate bad mols until I fix stereo issues~ Have validation separate out loaded molecules with problems, without modifying OFFTK.
  • [x] Add squelcher for RDKit stereo warnings
• [ ] Add more toolkit flexibility for setting conformers (allow deleting all/some)
• [ ] Need to intelligently select the "most diverse" confs from RDKitl while respecting user-supplied confs
• [ ] Wire up logging for this step and validation
• [x] ~Add error-dir input option to both conf gen and validation~ Add error dir in output dir

Status

• [ ] Ready for review

[1] Note to self: here's how to reproduce toolkit bug

from openforcefield.topology import Molecule
mol = Molecule.from_smiles('C=C')
mol.generate_conformers()
mol.to_file('xxx.sdf', file_format='sdf')
mol2 = Molecule.from_file('xxx.sdf')