Final changes for first official release

[x] Add links to datasets to readme
[x] Remove job id system
[x] Add ability to use single .h5 files as datasets, leveraging provided or automatically generated test/train indices. Any automatically generated indices should be saved in the model directory
[x] Allow the assess script to use indices
[x] Use batched computation in the assess script for efficiency
[x] Do not crop input randomly during assessment
[x] Move contents of main.py to main.py (allows python -m gerbilizer instead of -m gerbilizer.main
[x] Ensure default model parameters are consistent with hyperparameters used in the paper
[x] Prune some architectures from the main branch
[x] Allow ensemble "constituent configs" key to be filled with paths to config files.
[x] Remove gerbilizer.eval in favor of an inference option (suitable for performing evaluation without any ground truth) to gerbilizer.assess

neurostatslab / vocalocator