Totalsegmentator

* add independent script to download weights
* bugfixes

* support dicom input
* support dicom rt struct output
* add usage stats

* Correct wording in error messages
* add `--roi_subset` argument
* Use newer nnunet-customized version to avoid sklearn import error
* add `totalseg_import_weights` function
* add python api

* bugfix in cucim resampling
* add 6mm body model
* multilabel files contain label names in extended header
* add body model
* add pleural effusion model
* remove SimpleITK version requirement

* bugfixes
* add lung_vessels model
* add intracerebral hemorrhage model
* add coronary artery model
* preview file was renamed from `preview.png` to `preview_total.png`
* Split very big images into 3 parts and process one by one to avoid memory problems
* fix: check if input is 4d and then truncate to 3d
* make it work with windows
* make it work with cpu

* make output spacing exactly match input spacing
* improve weights download