allow_resume can be very slow

[jkosinski]

predict function of get_existing_model_info of predict_structure.py can be very slow on resume when there are many models/pkl files as it iterates and opens each pkl.

AlphaPulldown/alphapulldown/predict_structure.py

Line 126 in aaf63fc

ranking_confidences, unrelaxed_proteins, unrelaxed_pdbs, START = get_existing_model_info(output_dir, model_runners)

Possible improvements:

1. for finished jobs, check if ranking_debug.json and PAE pngs exist and do not re-read pkls in that case - just proceed to relax if requested, reading scores and filenames from ranking_debug.json or stop script if relax not requested
2. consider saving ranking_debug_temp.json with scores for models calculated so far