Load Xenium mask labels using Dask

[marcovarrone]

Towards #228

Replace the code to load the Xenium segmentation masks to use Dask to greatly reduce the memory consumption of the Xenium reader.

I abandoned the idea of saving each element to free up memory because the Xenium reader is already mostly Dask-compatible, so it will not bring any advantage.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 86.04651% with 6 lines in your changes missing coverage. Please review.
✅ Project coverage is 55.25%. Comparing base (5f829f5) to head (54a24a0).

Files with missing lines	Patch %	Lines
src/spatialdata_io/readers/xenium.py	86.04%	6 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #337      +/-   ##
==========================================
- Coverage   55.30%   55.25%   -0.05%     
==========================================
  Files          26       26              
  Lines        2846     2843       -3     
==========================================
- Hits         1574     1571       -3     
  Misses       1272     1272              

Files with missing lines	Coverage Δ
src/spatialdata_io/readers/xenium.py	72.70% <86.04%> (-0.25%)	⬇️

... and 2 files with indirect coverage changes

🚀 New features to boost your workflow:

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

[LucaMarconato]

Hi @marcovarrone, my apologies for the delay in reviewing this (or the other PR).

Great idea! In this PR the tests fail because the temporary directory where we extract the zip data is cleaned up when the function exits. I think opening the .zarr.zip file using a ZipStore should do the trick.

Meanwhile, to benchmark the performance gain in a more transparent way, I rely on an explicit extraction. I'll do some benchmarks and document the changes.