Major updates!

[mese79]

This is a major update over several parts (sorry for many changes)!
I needed to add an option for extracting image embeddings without patching, so I decided while adding this feature, make a major refactoring as well.

• Added a checkbox for no_patching option (only works for images with height == width)
  • With this option checked, the whole image (slice) will be treated as a single patch, and this speeds up the feature extraction and prediction (beneficial for images equal or smaller than 512x512).
• Made a dataset class: FFImageDataset
  • A torch IterableDataset that yields images / patches in batch.
  • The input image can be a numpy array, or a (large) stack file, or a directory of images.
  • Images will be lazy-loaded using pims (except for numpy array image, which already loaded).
  • This help unifying gui and pipeline script using the same set of functions.
• Now zarr is being used as the feature storage file (old storages are not compatible anymore, sorry).
  • Mainly because of appending features of zarr array, and more control over compression.
• run_pipeline.py can be used for large stack prediction on hpc without having a temporary feature store (faster). Also, it has an option for only extracting features into a zarr storage now.
• Default overlap is now patch_size // 4 as opposed to patch_size // 2 before. This will speed up extraction and prediction (I need to check the result stats with this change).
• Added more testing and typing.

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 63.68564% with 134 lines in your changes missing coverage. Please review.

Project coverage is 28.18%. Comparing base (5183db5) to head (d26ae91).

Files with missing lines	Patch %	Lines
src/featureforest/_segmentation_widget.py	36.02%	87 Missing ⚠️
src/featureforest/_feature_extractor_widget.py	17.39%	19 Missing ⚠️
src/featureforest/models/SAM/adapter.py	12.50%	7 Missing ⚠️
src/featureforest/models/MobileSAM/model.py	40.00%	3 Missing ⚠️
src/featureforest/models/SAM2/model.py	57.14%	3 Missing ⚠️
src/featureforest/utils/data.py	83.33%	3 Missing ⚠️
src/featureforest/utils/extract.py	93.18%	3 Missing ⚠️
src/featureforest/utils/pipeline_prediction.py	88.00%	3 Missing ⚠️
src/featureforest/utils/dataset.py	96.42%	2 Missing ⚠️
src/featureforest/models/Cellpose/adapter.py	50.00%	1 Missing ⚠️
... and 3 more

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #44      +/-   ##
==========================================
+ Coverage   24.85%   28.18%   +3.33%     
==========================================
  Files          40       40              
  Lines        2394     2469      +75     
==========================================
+ Hits          595      696     +101     
+ Misses       1799     1773      -26     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[mese79]

I need to update documentation as well.

[mese79]

Using zarr data storage seems to make RF training slow. :(
I would probably revert it back to HDF5.