Add QAA-AE model

[braceal]

Building off of the pyanca package detailed here: https://csb.pitt.edu/anca/index.html https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5961460/#mmc1 https://github.com/ChakraLabPitt/pyanca

We want to implement SD2, SD4, TD2, TD4, alignment functions in torch to take advantage of GPU acceleration. The outputs should easily plug into a AE model.

[braceal]

Prefer functions over classes. Alignment classes should just be functions.

[braceal]

Todo list:

• [x] Add sklearn like interface to extract loss curves - Collect a dictionary of train, valid loss. Return dictionary?
• [x] Fix plotting bug. Don't log checkpoint or plots on epoch 0. Should be epoch 1 not 0. epoch is 1 indexed.
• [x] Add inference function
• [x] Fix deterministic code issue
• [x] Add documentation for LinearAE model
• [x] Add docstring for LinearAETrainer class
• [x] Add documentation for LinearAETrainer
• [x] Add documentation for FeatureVectorDataset classes
• [x] Add alignment functions with documentation
• [ ] Add numpy implementations of SD2, SD4, TD2, TD4
• [ ] Add docstrings for SD2, SD4, TD2, TD4
• [ ] Add documentation for SD2, SD4, TD2, TD4
• [ ] Make jupyter notebook example with BBA for running SD2, SD4, TD2, TD4 + AE model
• [x] Remove unused err.T computation in kabsch function
• [ ] Ask Arvind for help with ANCA documentation
• [ ] update trainer prints to tqdm
• [x] update trainer script to raise TypeError if scalars is not a dict
• [ ] Add citation for QAA: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0015827

[braceal]

Merged support for the AE model into the main branch #9