perf: use set_virtual_refs_arr for faster icechunk writes

[TomNicholas]

Summary

• Replaces the Python loop that creates per-chunk VirtualChunkSpec objects with a single call to store.set_virtual_refs_arr(), passing the manifest's numpy arrays directly to Rust
• ~2.8x faster for writing virtual refs against object storage (12.4s → 4.5s for 6 vars × 520k chunks each against Arraylake)

Depends on earth-mover/icechunk#2049.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 73.06%. Comparing base (7a9e5fd) to head (9e13574).

❌ Your project check has failed because the head coverage (73.06%) is below the target coverage (75.00%). You can increase the head coverage or adjust the target coverage.

❗ There is a different number of reports uploaded between BASE (7a9e5fd) and HEAD (9e13574). Click for more details.

HEAD has 11 uploads less than BASE

Flag	BASE (7a9e5fd)	HEAD (9e13574)
unittests	12	1

Additional details and impacted files

@@             Coverage Diff             @@
##             main     #967       +/-   ##
===========================================
- Coverage   89.31%   73.06%   -16.26%     
===========================================
  Files          33       33               
  Lines        2031     2027        -4     
===========================================
- Hits         1814     1481      -333     
- Misses        217      546      +329     

Files with missing lines	Coverage Δ
virtualizarr/writers/icechunk.py	14.19% <ø> (-77.17%)	⬇️

... and 14 files with indirect coverage changes

🚀 New features to boost your workflow:

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