Adapt detection mask creation for Zarr/H5 files

[danifranco]

Now the detection masks are not created by chunks as it is done in the instance segmentation workflow. The whole image is loaded in memory and then dumped into the Zarr at the end.