Detect mismatched CF encoding attrs on concat

[TomNicholas]

Summary

• Fix silent data corruption when xr.concat (and any other path through np.concatenate on ManifestArray) is called on virtual datasets whose source files were CF-packed with different scale_factor / add_offset / _FillValue / missing_value. Previously only the first array's attrs survived, and the surviving values were applied to every chunk after decode_cf — silently mis-decoding everything sourced from non-first files.
• Route CF decoding attrs onto the inner ManifestArray.metadata.attributes (where np.concatenate dispatch can see them) rather than the wrapping xr.Variable.attrs; have check_combinable_zarr_arrays raise on any mismatch. Arbitrary attrs continue to live on xr.Variable.attrs.
• Writers (icechunk, kerchunk) gain a small helper extract_cf_encoding_attrs(var) so they can pull the CF attrs back out when materializing the destination store's metadata, keeping the round-trip identical.

Closes #1004.

Test plan

• New TestConcatMismatchedCFEncoding (in tests/test_parsers/test_hdf/test_hdf.py) builds real HDF5 files with mismatched scale_factor / add_offset and asserts xr.concat raises ValueError. Positive control with matching attrs concatenates cleanly. Committed first as a failing-by-design test so the diff makes the fix easy to read.
• Manual round-trip reproducer (HDF parse → xr.concat of matching files → to_icechunk → xr.open_zarr + decode_cf) produces correctly decoded values matching xr.open_dataset on the source.
• Full local suite passes: 485 tests + manifests, writers, parsers, xarray. Only pre-existing lithops failures (Python 3.11/3.13 runtime mismatch) skipped — confirmed on main too.
• CI

Notes for reviewers

• Relationship to the pre-existing check_compatible_encodings in writers/icechunk.py: that check fires only during to_icechunk(..., append_dim=...) and inspects var.encoding, which is the xarray-level slot CF attrs live in after decode_cf or when manually set with encoding={...}. For parser-produced virtual datasets (which never go through decode_cf) var.encoding is empty, so the existing check is a no-op on the most common real-world flow. The new check covers ManifestArray.metadata.attributes. They overlap (both fire on icechunk-append with mismatched parser data) but each covers a flow the other doesn't.
• to_virtual_variable now splits attrs by role. User-visible side effect: on a virtual variable, vds["foo"].attrs no longer contains _FillValue / scale_factor / etc. — they are available via vds["foo"].variable.data.metadata.attributes (and end up on var.attrs after writing + re-opening through xarray as normal).
• Kerchunk's MultiZarrToZarr.second_pass has the same bug pattern (see issue xr.concat of virtual datasets silently drops mismatched CF encoding (scale_factor, add_offset, _FillValue) #1004 "Related"). Worth flagging upstream too — not addressed here.

[codecov]

Codecov Report

❌ Patch coverage is 88.88889% with 3 lines in your changes missing coverage. Please review.
✅ Project coverage is 89.93%. Comparing base (ac60cf7) to head (7e812f8).
⚠️ Report is 11 commits behind head on main.

Files with missing lines	Patch %	Lines
virtualizarr/manifests/utils.py	85.71%	2 Missing ⚠️
virtualizarr/xarray.py	80.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1005      +/-   ##
==========================================
- Coverage   89.97%   89.93%   -0.04%     
==========================================
  Files          36       36              
  Lines        2224     2246      +22     
==========================================
+ Hits         2001     2020      +19     
- Misses        223      226       +3     

Files with missing lines	Coverage Δ
virtualizarr/manifests/array.py	85.98% <100.00%> (+0.13%)	⬆️
virtualizarr/writers/icechunk.py	91.71% <100.00%> (+0.04%)	⬆️
virtualizarr/writers/kerchunk.py	73.11% <100.00%> (+0.29%)	⬆️
virtualizarr/xarray.py	86.17% <80.00%> (-0.27%)	⬇️
virtualizarr/manifests/utils.py	91.02% <85.71%> (-1.17%)	⬇️

🚀 New features to boost your workflow:

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