One non-obvious advantage of the split is that some of the operator-counting constraints drag in large dependencies themselves (from memory, which may be wrong: things like PDBs and LM-Cut). So the price to pay (in terms of codebase size more than anything else) for including one specific OCC can be larger than is immediately apparent.

Currently, we can only enable/disable all operator-counting constraints as a group but it would be possible to put them in separate plugins with dependencies on a main plugin for the heuristic.