Losing molecules when converting torsiondrive datasets to optimizations

[jthorton]

When trying to convert a torsiondrive result class into a new restrained optimization dataset we lose some molecules and optimizations due to the chose of the index when a molecule is present more than once in a torsiondrive dataset due to it having multiple rotatable bonds. Maybe it would be best to change the conversion to use the original torsiondrive index and then tag it in the normal way with an index?

This could also simplify the construction of the dataset, currently, all restrained optimizations are separate entries in the dataset but we should get the same result by using all of the final geometries as starting geometries to an optimization and just freezing the dihedral angle. QCSubmit would then automatically generate the correct index for the dataset at submission time. This would lose the information on the intended torsion angle of the optimization as the constraint is more general but we could insert this into the attributes dict with the entry.