Mantis processing pipeline

[edyoshikun]

For future analysis of mantis datasets, the following order of steps will facilitate the processing:

1. Conversion to Zarr (light-sheet and label-free)
2. Label-free reconstruction
3. Deskew of fluorescence channels
4. Z and XY drift stabilization - register timepoints (typically in LF arm) to t=0
5. Registration - register moving dataset to t=0 of fixed dataset
   -- We need to VS one FOV for registration
   -- the registered datasets will have the voxel size of the fixed dataset
   -- fixed channel should be the one with lower spacial resolution such that the moving channel is resamples in that space
   -- if fixed and moving datasets have comparable resolution, the moving dataset should be the one with fewer channels such that less data is interpolated
6. Merging phase and fluorescence **
7. Virtual Staining - ideally append channels to registered datasets, see Need a pipeline to easily append new channels to existing stores #84
8. Segmentations - ditto
9. Analysis from segmentations

** Where should we be merging the datasets and if that is necessary at all (i.e step 5)?

The folder structure should follow something like below. Additionally, the VS steps always output the scale metadata as (1,1,1,1,1), which by default should be the input dataset's scale (here)

[ieivanov]

Thanks for creating this issue! I'll add a few notes later. Let's also create documentation for the structure of the analyzed datasets, similar to the structure of the raw datasets we have documented here: https://github.com/czbiohub-sf/mantis/blob/main/docs/data_structure.md