Threshold identification improvements
Some improvements to the double threshold implementation:
- Removed the array masks stored in the objects, using fields on the xarray dataset now to store the masks.
- Added the option to generate unique object ids in the postprocessing step (active by default).
- The identification class now provides the protobuf template for the output objects and the FeaturePipeline is now able to read this (backwards-compatible, the user can override this as before).
- Added docstrings and comments in the code.