Remove custom parallelisation in nlopt, use kpt_comm from calc instead

The nlopt module is parallelized over k-points; however, it currently needs the calculator to be initialized with serial_comm in order to use its custom parallelization scheme.

In this MR, we change the parallelization scheme to simply use the k-point communicator from the calculator. This is beneficial as it removes the necessity to save many GB of wave function data to disk in order to use the nlopt module.