As suggested in D5061, we don't really want to be using
STANDARD_SCRIPT_VERIFY_FLAGS for critical checks. In this case it's
not consensus validation-critical, but undercounting of sigops in ATMP
would lead to producing bad block templates (this was precisely the issue
in the May 2019 mining DoS attack).
This consolidates the sigops check into one place, also the comment
was a bit confusing (the per_mb limit is irrelevant).