Closes #4291.

Rationale for this change

Follow-up to #4292, which handled the rewrite itself. This covers the retry conflict check from the issue.

When a commit retries, _validate_no_new_deletes_for_data_files checks that no new delete file applies to the data files being replaced. It built its delete index with the operation's filter, though, so _filter_manifest_entries skipped any delete file whose column bounds can't match that filter. A partial delete rewrites those data files in full, so a concurrent delete outside the filter went unnoticed. The retry then committed the rewritten file, and the rows that delete removed came back.

Now every delete file added since the starting snapshot is checked against the replaced data files. The filter is no longer used there, so data_filter and case_sensitive are dropped from the private function's signature. _validate_no_new_delete_files, the filter-based check, is unchanged.

Are these changes tested?

Yes. test_partial_delete_validates_concurrent_delete_outside_the_filter in tests/table/test_commit_retry.py runs against each catalog fixture:

• a stale table handle starts delete("k = 1");
• meanwhile, a position delete storing the row (2, 'b') is committed. It has bounds k = 2, built with commit_partition_position_delete;
• the retried delete must raise ValidationException, leaving k = [1, 3].

Without the fix, the test fails with DID NOT RAISE ValidationException, because the retry commits over the concurrent delete.

Are there any user-facing changes?

A partial delete, filtered overwrite or upsert that races with a delete on the same data file now fails validation on retry, instead of silently bringing the deleted rows back.