[WIP] Add NR-NeRF method

[JulianKnodt]

NR-NeRF is another approach to dynamic NeRF that uses rigidity and additional losses.

I don't know if dedicating a whole set of configs to dynamic approaches is good, maybe it would be good to create a subcommand and have dynamic configs under that?

I can't test it rn because I'm on a machine that has 3 GB and it seems that some tensor is being stored that leads to an eventual OOM even if I only use 50 rays per batch

Edit: turns out eval was causing OOM not mem leaking from somewhere else