Forbid zero-length chunk references in ChunkManifest

[TomNicholas]

Closes #1088.

A chunk must decode to the full chunk shape, so no valid chunk is ever zero bytes long — even an all-constant compressed chunk carries a codec header, and arrays with a zero-size dimension have no chunks at all. A zero-length reference is therefore always a bug, but nothing rejected it until read time, by which point it had already been committed to a store:

InvalidInputError: byte range From(0) is out of bounds for chunk of length 0

This PR instead validates at construction. ChunkManifest is the choke point every ManifestArray routes through, so the check covers all three constructors (ChunkManifest, ChunkManifest.from_arrays, ChunkEntry.with_validation) and both virtual and inlined references. A chunk which simply isn't stored keeps its existing spelling — the empty path — which reads back as fill_value and fetches nothing.

ValueError: Found a zero-length chunk reference to 'file:///…/sparse.tif' at index (0, 1).
A chunk must decode to the full chunk shape, so no valid chunk is ever zero bytes long. To
record that a chunk is not stored at all (as in a sparse array), give it the empty path ''
instead, which reads back as the array's fill_value without fetching anything.

Performance

from_arrays scans lengths first and only compares the paths array if a zero length is actually present — the StringDType comparison is ~30x more expensive than the uint64 scan. At 10M chunks with validate_paths=False (~0.3 ms before this PR), a manifest with no missing chunks costs 1.6 ms and one with any missing chunks ~52 ms.

The check is deliberately not behind validate_paths, which guards path validation specifically — the expensive np.vectorize loop. Gating it would mean bulk-generating parsers, exactly where this class of bug arises, are the ones that lose the guard.

Also here

conftest.py was generating scalar arrays with length: 0, because the scalar branch passes chunks=(0,) and np.prod((0,)) * itemsize == 0. A scalar array's single chunk holds one element, so that's a fixture bug — fixed, with the two affected assertions updated.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 89.79%. Comparing base (8dc2cb7) to head (ca2baa9).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1089      +/-   ##
==========================================
+ Coverage   89.67%   89.79%   +0.12%     
==========================================
  Files          41       41              
  Lines        2682     2686       +4     
==========================================
+ Hits         2405     2412       +7     
+ Misses        277      274       -3     

Files with missing lines	Coverage Δ
virtualizarr/manifests/manifest.py	94.15% <100.00%> (+0.15%)	⬆️

... and 1 file with indirect coverage changes

🚀 New features to boost your workflow:

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