Using a `set` to lookup recipients that already received an email is much faster on average than with a list. `x in list` is O(n) on average, whereas `x in set` closer to O(1) on average. Example: for a sample mailing with 50k recipients that is half-through, filtering the remaining half (25k) took 10s with a list, and 1s with the set. X-original-commit: 7948698cc6e6b0af5fc84847ea7ec166882eda0b