Fix issue #506

[lohedges]

Closes #506. Sire removed the MM non-bonded interactions between QM atoms only when both atoms were in the same molecule, via that molecule's intrascale property. When the QM region spans several molecules and they react with each other, such as a water transferring a proton to a residue and its hydroxide bonding to a ligand, the full MM LJ between the newly bonded atoms remained and the simulation crashed. sire_to_openmm_system now adds an OpenMM::NonbondedForce exception for every pair of QM atoms in different molecules, plus matching exclusions in the ghost/ghost and ghost/non-ghost forces. The LambdaLever scales these from the full MM interaction at λ=0 to zero at λ=1 using the same clj charge_scale and lj_scale levers as the intramolecular QM pairs, so λ interpolation stays consistent. To do this, QMForce now exposes a pure virtual getAtoms(), which PyQMForce and TorchQMForce already implement. A regression test, test_qm_cross_molecule_exceptions, checks that the exceptions exist and take the right values at both end states.

• I confirm that I have merged the latest version of devel into this branch before issuing this pull request (e.g. by running git pull origin devel): [y/n]
• I confirm that I have added a changelog entry to the changelog (we will add a link to this PR as part of the review): [y/n]
• I confirm that I have permission to release this code under the GPL3 license: [y/n]