Log charge provenance once per unique molecule

[mattwthompson]

Closes #1484
Closes #1483

(I found this was easier to make these two changes in one go, but I can refactor it into separate changes if desired)

Description

With these changes:

• The charge method used to apply partial charges to a given molecule is only logged once
  • Previously this was per-atom, and therefore repetitive. Now it's just once per molecule
• The charge method used to apply partial charges to a duplicates of a unique molecule is only logged once
  • Previously, this was duplicated for each duplicate molecule, so e.g. 1000 waters would be 3000 lines in the log
• InChI is used to identify the moleule
  • SMILES for all molecules won't work, because of proteins
  • Another solution could be to use SMILES for small molecules like water and ions but InChI for molecules above some threshold size
  • Molecule.to_inchi failures are handled with a wide net
• Logging logic is not entered unless it's turned to logging.INFO

Here is an example of it in use, adopting from others' code

In [1]: import logging
   ...: logging.basicConfig(level=logging.INFO)
   ...:
   ...: from openff.toolkit import ForceField, Molecule
   ...: from openff.interchange import Interchange
   ...: from openff.interchange.components._packmol import pack_box
   ...: from openff.units import unit
   ...:
   ...: mols = [Molecule.from_smiles(smiles) for smiles in ['CCO', 'O']]
   ...: for mol in mols:
   ...:     mol.generate_conformers(n_conformers=1)
   ...:
   ...: topology = pack_box(
   ...:     molecules=mols,
   ...:     number_of_copies=[3, 2],
   ...:     target_density=0.1 * unit.gram / unit.milliliter,
   ...: )
   ...:
   ...: ff = ForceField("openff-2.2.1.offxml")
   ...: interchange = Interchange.from_smirnoff(force_field=ff, topology=topology)
WARNING:pint.util:Redefining '[electric_potential]' (<class 'pint.delegates.txt_defparser.plain.DerivedDimensionDefinition'>)
/Users/mattthompson/software/openff-interchange/openff/interchange/components/interchange.py:1119: FutureWarning: `torch.distributed.reduce_op` is deprecated, please use `torch.distributed.ReduceOp` instead
  if isinstance(obj, functools._lru_cache_wrapper) and obj.__module__.startswith("openff.interchange"):
INFO:openff.interchange.smirnoff._nonbonded:Charge section ToolkitAM1BCC, using charge method am1bccelf10, applied to molecule with InChI InChI=1S/C2H6O/c1-2-3/h3H,2H2,1H3
INFO:openff.interchange.smirnoff._nonbonded:Charge section LibraryCharges applied to molecule with InChI InChI=1S/H2O/h1H2

I have not done it yet, but, in principle, a large solvated protein-ligand system may have its entire charge provenance logged with as few as 5 lines i the log

Checklist

• Update tests
• Lint
• Update docstrings

[codecov]

Codecov Report

❌ Patch coverage is 84.61538% with 4 lines in your changes missing coverage. Please review.
✅ Project coverage is 94.05%. Comparing base (14cd931) to head (33a1911).
⚠️ Report is 2 commits behind head on main.

Files with missing lines	Patch %	Lines
openff/interchange/smirnoff/_nonbonded.py	84.61%	4 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1488      +/-   ##
==========================================
- Coverage   94.07%   94.05%   -0.03%     
==========================================
  Files          73       73              
  Lines        6045     6056      +11     
==========================================
+ Hits         5687     5696       +9     
- Misses        358      360       +2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[mattwthompson]

This is a subtle behavior change.

Previously, there was logging for the charge method applied to the whole molecule (partial_charge_method in the SMIRNOFF data and API) and then also each individual charge increment. Each of these are done on a "per-atom" view, if repetitive, same as how other charge methods were logged before these changes.

Taking a "per-molecule" view works nicely for other methods (no need to log $N_{atoms}$ lines for every atom in molecule that's given AM1-BCC charges) and works fine for ChargeIncrementModelHandler.partial_charge_method but doesn't work for each individual charge increment (i.e. BCC). There's probably a way to refactor all of this to cater to that detail, but I don't think it's trivial. Given that ChargeIncrementModel is not actually used in production, and there aren't plans to do so, I'm inclined to not put that effort into this change.