Refactor prediction pipeline

[jdeschamps]

Description

tldr:

• Split prediction datasets into tiled and non-tiled
• Simplify stitching by passing TileInformation all along
• Fix Stitched prediction not compatible with multi channel [BUG] #125
• Change total memory check to available

When not tiling prediction, the images are forced through the tiling pipeline (with a TileInformation class being passed along the images), and it makes the debugging complex. This PR splits the prediction datasets into tiled and not-tiled.

I also changed the total memory into available for the switch between in memory and iterable datasets during training, as it better represents what can be loaded in memory.

Finally, I simplified the stitching and prediction pipeline by passing the TileInformation further.

• What: Refactor prediction datasets into tiled and non-tiled datasets, simplify stitching.
• Why: Avoids forcing non-tiled prediction through the same complex pipeline as the tiled one.
• How: Split the two features (tiled and non tiled predictions) into two datasets.

Changes Made

• Added:
  • dataset/iterable_pred_dataset.py
  • dataset/iterable_tiled_pred_dataset.py
  • dataset/in_memory_pred_dataset.py
  • dataset/in_memory_tiled_pred_dataset.py
• Modified: get_ram_size now looks at available memory.
• Removed: Removed useless calls to sort in datasets, is_tile in TilingInformation.

Related issues

This PR fixes #125.

Notes

This PR will create merge issues with #134.

Currently, there are two issues remaining:

• extract_tile returns C(Z)YX, while the non tiling datasets always return SC(Z)YX with a singleton dimension
• it is not clear when we should, and when, cast the Tensors into np.ndarray. I tried to make it happen in the same place (in the prediction loop), but that is not entirely solved.

---

Please ensure your PR meets the following requirements:

• Code builds and passes tests locally, including doctests
• New tests have been added (for bug fixes/features)
• Pre-commit passes
• PR to the documentation exists (for bug fixes / features)

[codecov]

Codecov Report

Attention: Patch coverage is 95.92760% with 9 lines in your changes missing coverage. Please review.

Project coverage is 91.33%. Comparing base (4767329) to head (616fac0).
Report is 69 commits behind head on main.

Files	Patch %	Lines
...eamics/dataset/dataset_utils/iterate_over_files.py	92.85%	2 Missing ⚠️
.../careamics/dataset/in_memory_tiled_pred_dataset.py	94.59%	2 Missing ⚠️
...c/careamics/dataset/iterable_tiled_pred_dataset.py	93.93%	2 Missing ⚠️
src/careamics/config/tile_information.py	83.33%	1 Missing ⚠️
src/careamics/dataset/iterable_pred_dataset.py	96.15%	1 Missing ⚠️
src/careamics/lightning_prediction_datamodule.py	90.90%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #131      +/-   ##
==========================================
+ Coverage   81.90%   91.33%   +9.42%     
==========================================
  Files         103      104       +1     
  Lines        2841     2561     -280     
==========================================
+ Hits         2327     2339      +12     
+ Misses        514      222     -292     

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