Molalkit

support active learning that assign active set and validation set in two files.

support partial query data, which is efficient for large-scale dataset.
support --output_details to save the per-sample predictions on the validation set during active learning.

Support terminating active learning when the acquisition values exceed a certain cutoff (stop_cutoff).

Add unittest.
Bugfix for forgetters.
updates model configs.

First release of MolALKit.