Fix form of null bonded terms

[lohedges]

This PR fixes the form of null bonded terms used during the somd1 compatibility mode. By not adding the terms in the end state in which they should be nulled, sire will correctly add a null term that zeros the force constant but preserves the other components, e.g. r0, theta0, phase, periodicity, from the opposite state. When adding an explicit null term, conversion to a CAS expression creates a fully nulled potential, so all components are zero, i..e. a bond would have its force constant and bond size perturbed to zero. I don't think this causes issues in practice, since it's just a different pathway through lambda space, but it is good to be consistent with what is done for somd1.

[lohedges]

Just to add that this doesn't solve the ethane<-->methanol reversibility problem ;-)

[mb2055]

Looks good 👍 Well spotted

[lohedges]

I'll hold off on this for a sec. From the emle side I am finding that I sometimes get errors such as:

Have a 1-4 scaling factor (0.833333/0.5) between atoms N:6 and HH31:0 despite there being no physical dihedral between these two atoms. All
1-4 scaling factors MUST be associated with physical dihedrals. The shortest path is [  ]

when dihedrals are removed from an end state. I would have assumed that a null would be added so that the term would be there.

[lohedges]

Thanks, I just want to double check that nulling terms in this way doesn't cause issues for dihedrals that don't only contain ghost atoms. This was what was causing my issue with sire-emle above.