BUG: write_csv fails midway and leaves a partial file when later rows carry extra keys #3303

Description

write_csv in pyrit.common.csv_helper builds the CSV header from the keys of the first row only. Any later row carrying a key the first row lacks makes csv.DictWriter.writerows raise ValueError: dict contains fields not in fieldnames, and because the exception happens after the header and earlier rows were already written, the caller is left with a truncated partial file on disk.

This matters for the remote dataset cache path, which writes fetched rows through this helper: dataset rows can legitimately vary in which optional fields they carry, and a single wider row turns a cache refresh into a corrupt cache file.

Expected behavior

The header should be the union of keys across all rows in first-seen order, and rows missing a header key should get an empty cell instead of an error.