Added optional additional rejection criteria of walkers based on nearest neighbour distance

[VGFletcher]

When using machine learned potentials they often predict nonsensical energies at very small interatomic distances. I've been using a rejection criteria to prevent these unphysical configurations ruining a nested sampling run. The user supplies a minimum nearest neighbour distance (min_nn_dis) and if an initial walker is generated with a nearest neighbour distance less than this value, the walker is rejected. The criteria only applies to the initial generation but can be applied to the entire run by setting calc_nn_dis to true. The rejection criteria, and associated calculation, is optionally enabled and doesn't impact a normal run. I added comments and an explanation of the keywords for documentation.

I also formatted the intro.html doc page from "MC and MD step algorithms" onwards to make it a bit easier to read, I didn't change any content. The page that's up currently on the doc page is broken but when I initially forked and compiled without any modification it removed the warnings so I think that page just needs to be recompiled. I can remove my formatting if it is not welcome.

Best regards,
Vincent.

[bernstei]

I'll let @liviabp comment on this, since I've completely deprecated pymatnest in favor of pymatnext