Improve parallelisation of OME-Zarr writing

[jluethi]

Currently parallelisation is by well. For the test case /data/active/fractal/3D/PelkmansLab/CardiacMultiplexing/Cycle1_subset/, that results in ~30 minute run time for a single well. Thus, options to further parallelise this would be very helpful for small to medium datasets as well as to being able to run these kind of test cases better.

See here for original parallelisation discussions: #16

Given the zarr architecture, well should be the primary split point. Then time point would be an option, though we often only have 1 timepoint in current datasets (but once there are more, parallelizing over them would make sense). After that, there are the pyramids, so parallelization below the pyramid level may be a bit more complicated. Still, may make sense to parallelize by channel or z plane after that.

[jluethi]

See here for the discussion on zarr writing task efficiency & memory usage: #57 (comment)

[jluethi]

Zarr writing performance is very good overall and improved with the dask parallelization here fractal-analytics-platform/fractal-tasks-core#20. No need to have Fractal distribute this further (and would anyway be hard/impossible with the single FOV per well strategy).
Closing this for no.