Add docker

[decarboxy]

This PR addresses issue #14 and takes a somewhat different approach from the dockerfile posted in #54. This dockerfile aims to be relatively compact (It comes out to about 8 GB which is large for a docker image but about the same size as the image produced by the alphafold repository).

As is the best practice for docker, the model files, sequence data, and structural data used by openfold are not included as part of the model, rather, the user should mount directories containing that data as volumes. the Readme file has been updated with example commands for building and running the container.

In addition to the dockerfile, I made a change to openfold.np.resources.load_stereo_chemical_props to load the stereochemical properties using importlib.resources rather than a relative path, this allows the user of the code to call it from any working directory.

One odd thing I ran into was that it was necessary to install openCL in addition to cuda in order for the relax phase of the modeling pipeline to work. I haven't looked into why openMM was refusing to use the cuda platform and insisted on openCL. That said, I tested this and it does seem to work fine using the alphafold weights.

[gahdritz]

Thanks! The OpenCL thing is the result of a hack in run_pretrained_openfold.py. I've been meaning to remove it, but for now it's still there.

[gahdritz]

For the record: I've removed the OpenCL dependency. It appears to not be necessary anymore, but if the relaxation script starts hanging at any point, this is probably why.