MMFF94: environment-dependent q⁰ for the O2CM terminal oxygen (type 107) — fixes spurious net charge on sulfones and related groups

[exergonic]

While validating our own port of the force field against reference implementations, we noticed the net charge was off by a full electron on some structures — dimethylsulfone came out at −1.00 e, methanesulfonate at −1.50 e. Traced it to kchargem: type-107 (O2CM) always gets a flat base charge of −0.5 e, but that type is shared by a lot more than one functional group. MMFF94's class 32 covers the terminal oxygens of sulfones, sulfonates, sulfates, nitro groups, N-oxides, phosphine oxides, and carboxylates, and the class→internal mapping sends all of them to 107. A flat charge only works if every one of those centers has the same number of terminal oxygens relative to its neutral parent oxyacid — they don't. Per part V eq. (15) of Halgren's paper, q⁰ should be −(n−k)/n, where n is the terminal oxygen count and k depends on the central atom (1 for C/P/S(IV), 2 for N/S(VI), 3 for Cl). With the flat value wrong, minimization follows a distorted surface. One sulfonamide came out 37.7 kcal/mol below its correct minimum.

This fix computes q⁰ for type 107 from the local environment using that formula instead of the constant. Centers outside the rule are untouched, as are sibling types (109, 113, …). Checked against RDKit's independent MMFF94 implementation on 12 compounds: all nets now correct, per-atom charges agree to ≤3×10⁻⁵ e, energies match to print precision, and existing test controls stay bit-identical.

Addresses #185.