Mimsi

* Added MIL attention pooling as an option for increasing the interpretability of the model
* Updated models after further training
* Additional tests added
* Updated pytorch
* Documentation clean up

* Force coverage param in Model instantiation to integer rather then string

* Add latest version of the lower coverage (100x combined tumor/normal) model
* fix to address coverage parameter mistakenly interpreted as string

- Hotfix to address some missing arguments for the last validation testing call in our custom training script

Merging in the latest MiMSI update, includes:

- fixes and speed improvements to the bam2tensor model
- removal of the giant microsatellites.list file in favor of a smaller one limited to only IMPACT specific regions (significantly speeds up data generation from bam files)
- training/testing improvements and bug fixes
- updated pre-trained model

- Appease the linter
- Modifying Case list format to include column names
- Add Normal ID as an option to the case list, this is helpful in the event we need to record the normal used in MSI calling
- travis CI updates and py3 cleanup