Fix bilinear cache type decoding

[Zaczero]

Fixes bilinear cache type decoding:

mask_slices are boolean mask, and are stored as int8 via zarr.

boolean information is part of xarray metadata, so when skipping it we get back int8.

now mask_slices is not a mask but rather an array of 0, 1 indices making the mask useless.

---

I opted into chunks=auto automatic rechunking so Dask can pick the optimal chunk size, which gave a small performance boost on my data when using the cache:

  - raw zarr + astype(bool): load avg 0.006923s, compute avg 0.020060s
  - xarray open_zarr(chunks='auto'): load avg 0.004082s, compute avg 0.019647s

• Tests added
• Tests passed

[codecov]

Codecov Report

❌ Patch coverage is 80.00000% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 93.67%. Comparing base (4da28b0) to head (a5192ed).
⚠️ Report is 37 commits behind head on main.

Files with missing lines	Patch %	Lines
pyresample/bilinear/xarr.py	75.00%	1 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #702   +/-   ##
=======================================
  Coverage   93.67%   93.67%           
=======================================
  Files          89       89           
  Lines       13621    13656   +35     
=======================================
+ Hits        12759    12792   +33     
- Misses        862      864    +2     

Flag	Coverage Δ
unittests	93.67% <80.00%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

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

🚀 New features to boost your workflow:

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

[Manny7717]

Review — Approved

Verified locally (shapely 2.1.2, zarr installed): test_save_and_load_bil_info passes, including the new mask_slices.dtype == bool assertion.

Root cause is right: zarr stores Python bools as uint8 without dtype metadata, so da.array(fid[val]) re-exposed the raw uint8. open_zarr(...).data goes through xarray's dtype decoding and restores bool. The removed import zarr is safe — the only remaining zarr references in the module are the zarr_out variable and xarray's Dataset.to_zarr() / open_zarr, which lazy-import zarr themselves.

Non-blocking note (inline): caches written by the previous version will still decode as uint8 masks under the new load path (no dtype marker was stored). Harmless in practice since caches regenerate, but a comment in the changelog or a cache-format version bump would save someone a confusing debug session.

[Manny7717]

Minor: caches written before this change (raw uint8 bools via zarr.open) will still come back as uint8 under open_zarr, since no dtype marker was persisted. Caches regenerate on the next save, so it's benign — a changelog note (or cache version bump) would prevent future confusion.