Linked initialisation can be slow even when restoring response matrices from disk
The routine init_connected_bc currently involves several loops over the entire global domain. This means that the cost does not reduce as the processor count is increased and does increase as the problem size is increased. As such, this can come to dominate the initialisation costs when restarting a run.
It may be possible to convert these loops to loops over the local domain as was done previously with most of redistribute initialisation or perhaps possible to change these loops to instead loop over just the theta0/ky domains etc. (as the communication pattern involves different theta0/kx communicating at fixed ky, lambda, energy and species).
We may also be able to split this routine into several smaller routines as it currently calculates a few different quantities (such as for the field calculations) in addition to the linked boundary communication pattern.
