[BUG] Crystallographic waters could be removed during solvation

[lohedges]

I came across this potential problem when reading this issue thread. For OpenMM, existing crystallographic waters are (now) correctly preserved when adding ions. There is no logic do deal with this when using gmx genion, other than specifying the minimum distance between ions and non-solvent with -rmin, which doesn't seem to work very well. Although I imagine that this issue would rarely occur, we might need to add some logic to check which waters were replaced, and re-run if any crystallographic waters were removed. (Alternatively, we could just switch to using openmm.Modeller if it's more reliable.)

[lohedges]

This could be solved by having a a different water naming template for crystal waters, i.e. convert to that before solvating and using genion, then convert to the standard template afterwards.