Cache the expensive only_us_ascii? check - #1119

@jeremy Great using String#ascii_only?! The only difference is that the original regex excludes "\x00" while ascii_only? includes it, does it matter? With this change the test suite time goes from 2.3 seconds to 1.6 seconds. And also, I'm seeing significant performance improvements in benchmarks locally building real emails.

30 attachments of ~ 10MB each:   17.801965 sec -> 5.149144 sec (~ 3x faster)
10000 html parts of ~ 40KB each: 16.148903 sec -> 9.189533 sec (~ 1x faster)

IMO that was a bug, since we don't want NUL bytes in our US-ASCII text either :)

Very nice improvements!

It's the other way around, the new ascii_only? check will allow NUL bytes in emails I'm afraid. :) If we want to prevent that, we could do raw_source.ascii_only? && raw_source !~ /\x00/ :/, but that would be slower than ascii_only? and faster than the original regex.

Hmm yes. This seems like an implementation wart, really. NUL is legit for US-ASCII. It's not legal in emails, but that's not what we're validating here, so we can get away with the ascii_only?check.

Great, thanks for checking that, Jeremy.