perf: optimize arrays_zip_inner in datafusion-comet-spark-expr

[andygrove]

This PR was created by an LLM as a draft PR. I will mark it as ready for review after human review.

Which issue does this PR close?

N/A — autonomous exploratory PR.

Rationale for this change

Borrow list offsets/null buffers instead of materializing per-column Vecs, size the MutableArrayData builders exactly, and coalesce unpadded rows into bulk extends so equal-length rows cost one copy per column instead of one call per row.

What changes are included in this PR?

Borrow list offsets/null buffers instead of materializing per-column Vecs, size the MutableArrayData builders exactly, and coalesce unpadded rows into bulk extends so equal-length rows cost one copy per column instead of one call per row.

How are these changes tested?

Correctness: unit tests + seeded differential fuzz (bit-identical Arrow output vs main).

Benchmark (criterion):

• arrays_zip_uniform: 20.538% faster (base 362344ns -> cand 287927ns)
• arrays_zip_ragged: 11.864% faster (base 421085ns -> cand 371129ns)

Full criterion output:

arrays_zip/uniform      time:   [287.27 µs 287.91 µs 288.91 µs]
                        change: [−20.756% −20.538% −20.270%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 8 outliers among 100 measurements (8.00%)
  3 (3.00%) high mild
  5 (5.00%) high severe

arrays_zip/ragged       time:   [370.88 µs 371.03 µs 371.19 µs]
                        change: [−12.181% −11.864% −11.603%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 6 outliers among 100 measurements (6.00%)
  1 (1.00%) low severe
  1 (1.00%) high mild
  4 (4.00%) high severe