Improve charge provenance logging with large molecule(s)

[mattwthompson]

As @IAlibay and others have mentioned, charge provenance logging is more or less useless with large molecules like proteins. The original design logs each atom on a new line, so i.e. a protein given AshGC charges will log several thousand approximately-identical lines.

Irfan also suggests a solution: simply log the entire molecule in one line.

[mattwthompson]

A crucial technical detail that I didn't have years ago is that any molecule will only ever have one partial charge assignment method used with in it¹. I previously thought that some edge cases with library charges in proteins could mix-and-match two methods, but this is non-standard behavior. So, if we can capture all of the necessary provenance of partial charge assignment in a given molecule just by listing the method (and maybe a little extra detail), it could end up simpler like

<TIMESTAMP> DEBUG     Charge section NAGLCharges, using NAGL model openff-gnn-am1bcc-1.0.0.pt, applied to molecule 0 with InChIKey BLAHBLAH
<TIMESTAMP> DEBUG     Charge section NAGLCharges, using NAGL model openff-gnn-am1bcc-1.0.0.pt, applied to molecule 1 with InChIKey LFQSCWFLJHTTHZ-UHFFFAOYSA-N-UHFFFAOYSA-N
<TIMESTAMP> DEBUG     Charge section LibraryCharges, applied to molecule 1 with InChIKey XLYOFNOQVPJJNP-UHFFFAOYSA-N
<TIMESTAMP> DEBUG     Charge section LibraryCharges, applied to molecule 1 with InChIKey FKNQFGJONOIPTF-UHFFFAOYSA-N
-UHFFFAOYSA-N
<TIMESTAMP> DEBUG     Charge section LibraryCharges, applied to molecule 1 with InChIKey VEXZGXHMUGYJMC-UHFFFAOYSA-M

This is a faux protein-ligand system with only five topologically identical components (protein, ligand, water, sodium, chloride). I think this is about all the information a user wants, but drastically more compressed than current behavior (parent issue #1479)? I'm not sure how to handle molecule indexing; if a user wants to go from molecule index (i.e. from the 0th molecule in the topology to the charge method applied) there'd be a little extra work.

Note that de-duplicating copies of identical molecules is also under discussion #1483. I'd like to keep these separated as much as possible, although the example above relies on a sort of solution to that.

Footnotes

1. The case of <ChargeIncrementModel> is a little more tricky, since there can be both a semi-empirical method applied everywhere and individual charge increments (like BCCs) applied in the molecule. This is rare (not used in production) and for our purposes I'm just going to hand-wave and say "well, it's sorta two steps but it's all coming from one handler." ↩