Fix multiprocessing of structural featurizers

[schallerdavid]

Description

@corey-taylor discovered a bug leading to never finishing processes when using multi-processing in the structural featurizers without specifying an output directory. In this case, the generated openeye molecule object will be written to disk as a PDB file, read back in as MDAnalysis universe, the file will be deleted and finally, the universe returned. This is the intended behavior when not specifying an output directory since the user was not interested in the file. However, standard MDAnalysis readers do not load into memory. Hence, during serialization in the multiprocessing scheme, the MDAnalysis reader will miss the file, since it has been deleted.

This PR makes sure we read the generated structure into memory. Thus, we can delete the file and still serialize the universe.

Questions

• [ ] @corey-taylor Can you check this out and run the docking notebook with this branch?

Status

• [ ] Ready to go

[codecov-commenter]

Codecov Report

Merging #92 (dfdb17c) into master (5e38b6b) will not change coverage. The diff coverage is 25.00%.

[corey-taylor]

Have run test cases of docking with and without multiprocessing option - all worked smoothly.