Computing LFM requires memory and processing optimization

[neuromechanist]

Going through the BEM forward problem (see #6 for the FEM problem) is a multi-layer process that is neatly laid out in the GUI and works pretty well (on Linux and Mac at least, see #5), but computing LFM is largely lacking parallel processing, and when it does use the parpool, it seems that there are loops with very high memory usage. About 50% of the KFM processing is done with single tread, and multi-thread processing uses about 4 GB of RAM per thread. This must be < 2 GB for sustainable processing on a local workstation or HPC.