Use a ChimeraX fork of the Boltz github repository to improve Boltz behavior

[Tom Goddard]

It might be useful to have ChimeraX Boltz install use a fork of the Boltz repository that I maintain so that custom enhancements can be included. An immediately useful enhancement would be to add a --use_cuda_bfloat16 option to allow predicting much larger structures when Nvidia graphics is available. I can probably get the main Boltz repository to include that, but hard to tell how long it might take.

Colabfold uses a custom AlphaFold 2 fork.

There is the drawback that I can't easily get the Boltz code fixes from the main repository without merging and testing in some fork that will not effect ChimeraX users (in case things go wrong).

Maybe I should wait until I run into a situation where I can't get the main Boltz repository to accept changes I want to add.

Currently I am using the PyPi Boltz distribution (all Python source) and I may want to switch to installing from github since the current PyPi Boltz 1.0 is broken on non-nvidia systems but has been fixed in the github code.