fix: stop calling deprecated PuffinFile.to_vector in Spark DV test

[daviddallakyan2005]

Closes #3803

Rationale for this change

Follow-up to #3476. test_read_spark_written_puffin_dv still calls PuffinFile.to_vector(), which #3491 deprecated in 0.12.0. filterwarnings = ["error"] turns that into a hard CI failure, so integration-test is red on main and on unrelated PRs including #3801.

This test now uses deletion_vectors_from_puffin_file(...), matching pyiceberg/io/pyarrow.py. Spark interop assertions are unchanged. Not part of #3801.

Are these changes tested?

Yes. make test-integration PYTEST_ARGS="-v -k test_read_spark_written_puffin_dv" → 1 passed, 5517 deselected.

Are there any user-facing changes?

No.

[moomindani]

Thanks for catching and fixing this, @daviddallakyan2005, and sorry for the noise — the broken call came in with #3476.

For the record on the root cause: #3476's last commit was 2026-06-18, and #3491 deprecated PuffinFile.to_vector() on 2026-06-25. The test was green when it last ran and only went stale while the PR waited for a merge, which landed on 08-15 without a rebase — so CI never ran it against the current main.

The fix looks right to me. DeletionVector.to_vector() isn't deprecated, and the assertions are equivalent since the old loop never used the data file path key. I also checked that test_deletes.py:1084 was the only remaining call site of the deprecated method on main.

[ebyhr]

This is simply a logical conflict. No need to apologize :)

We could consider enabling the merge queue so that we can re-run CI just before its merge. Here's a reference:

• feat(github): add merge_queue convenience syntax for rulesets infrastructure-asfyaml#119