Darwin-py

- New `ml` extra for ML-related dependencies (`pip install darwin-py[ml]`)

- Deconstruct Paths for COCO
- fixed check on filename not existing (PascalVOC import issue)
- Add DARWIN_UPLOAD_CHUNK_SIZE environment variable (fixing uploads failing when uploads are too big)
- Improved failed requests handling
- Added try except for symlink creation (Now symlink creation should not crash in any platform)
- Add class to dataset based on annotation class id (adaption to new API)

- [HOTFIX] Convert target to Tensor in ClassificationDataset
- [FIX] Change default behaviour in ClassificationDataset
- Include progress when fetching dataset
- Upload files with PUT instead of POST requests (to conform with API updates)

- FIX: include sub-annotation data for all annotations converted to COCO
- ClassificationDataset in `torch` module is now usable for multi-label classification purposes

- import_annotations now raise when parameters have invalid format

- Avoid creating `latest` symlinks of dataset versions when pulling datasets