Speed up coh_tmm in tmm_core_vec; introduced versions with cpu and gpu parallelization

[griddler-j]

s and p polarization, 10000 wavelength x angles, 6 layers, calculate coh_tmm:
before speed increase: 4.178s
after speed increase: 2.399s
after speed increase, non-detailed mode: 1.948s
CPU parallelization with 24 cores
CPU parallelization: 0.844s
CPU parallelization, non-detailed mode: 0.719s
GPU parallelization with NVIDIA GeForce RTX 4060
GPU parallelization: 0.296s
GPU parallelization, non-detailed mode: 0.118s

[phoebe-p]

sorry for my lack of input on this. I was wondering, for the parallelisation, what do you think the best way to incorporate this would be? I guess there are now three options: no parallelisation (i.e. just the old implementation), GPU parallelisation, or CPU parallelisation. Then there's the detailed vs. non-detailed mode. The user should be able to choose which one they want to use, but it would be better not to have three different files tmm_core_vec files, since most of the content is the same anyway.