fix: preserve shard shape when writing to icechunk

[TomNicholas]

Summary

• Fixes the icechunk writer using metadata.chunks (inner chunk shape) instead of metadata.chunk_grid.chunk_shape (shard shape) when creating arrays, causing sharded arrays to lose their shard configuration
• Adds a regression test that creates a sharded zarr store, virtualizes it, writes to icechunk, and verifies both shard and chunk shapes are preserved

Closes #951

Test plan

• New test test_sharded_array_roundtrip_icechunk passes
• Existing icechunk writer tests unaffected (pre-existing test_write_loadable_variable failure is unrelated)

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 89.30%. Comparing base (fd2665c) to head (758c1ef).
⚠️ Report is 52 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #952      +/-   ##
==========================================
- Coverage   89.31%   89.30%   -0.02%     
==========================================
  Files          33       33              
  Lines        2041     2039       -2     
==========================================
- Hits         1823     1821       -2     
  Misses        218      218              

Files with missing lines	Coverage Δ
virtualizarr/writers/icechunk.py	91.35% <100.00%> (-0.11%)	⬇️

🚀 New features to boost your workflow:

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

[maxrjones]

@TomNicholas the current change will lose the inner chunk structure. I cannot drop in-line suggestion to L517, but I think this code block should look like:

        filters, serializer, compressors = extract_codecs(metadata.inner_codecs)
        arr = group.require_array(
            name=name,
            shape=metadata.shape,
            chunks=metadata.chunks,
            shards=metadata.shards,
            dtype=metadata.data_type.to_native_dtype(),
            filters=filters,
            compressors=compressors,
            serializer=serializer,
            dimension_names=var.dims,
            fill_value=metadata.fill_value,
        )

• metadata.chunks / metadata.shards tells require_array both the inner chunk shape and shard shape, so it constructs the ShardingCodec correctly
• metadata.inner_codecs — extracts the codecs inside the sharding codec (e.g. BytesCodec, ZstdCodec)

[TomNicholas]

oh interesting - thanks. Apparently this seemed to work even without this change, due to some subtlety in zarr-python? I/Claude couldn't find a way to make a test that failed before adding your suggestion but passed after.

But I'm made this change now anyway as it's much more explicit.