How it works: the worker auto-detects the email column,
resolves each domain's MX host, and paces probes per MX (~4s for small
hosts, ~6s for Google/Microsoft) with adaptive backoff on deferrals.
Rows marked invalid are dropped; everything else (safe / risky
/ unknown) is kept. No fixed daily cap — pacing and the Spamhaus check
are the only governors. Output rows are in completion order, not input
order; larger jobs spill across UTC days as separate downloadable CSVs.