Fixed instance seg dtype, allowed change of normalization quantile, added model checkpoint version control

[JLrumberger]

What is the purpose of this PR?

1. Add support for float instance segmentation maps as requested in 32bit images for cell segmentation as label mask #18
2. Add functionality to change normalization quantile and clip marker expression after normalization to fix Nimbus doesn't produce accurate protein expression in some area #17
3. Add functionality to automatically update model weights when we upload new checkpoints to the huggingface hub.

How did you implement your changes

1. Cast all instance segmentation maps to uint64 prior to further analysis.
2. Added normalization quantile and clipping to the interface of Nimbus.prepare_normalization_dict
3. Changed Nimbus.initialize_model to first check if there exists a checkpoint with a higher version tag ('V1.pt', 'V2.pt',...) on the huggingface hub. Then download the checkpoint and load the weights into the model.

Remaining issues

None.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[ngreenwald]

Looks good!