Possibly update charge symmetrization approach

[maxentile]

Migrated from https://github.com/proteneer/timemachine/pull/738

At least four approaches possible:

• Current approach -- stereo unaware, but occasionally introducing very small asymmetries in cases like benzene
• Initial approach proposed in #738 -- refinement of current approach -- still average over stereo-unaware atom classes, but reduce number of distinct partial charges assigned to benzene from 3 to 2
• Final approach proposed in #738 -- departure from current approach, average over stereo-sensitive atom classes
• No symmetrization