Slow IDNA encoding with large strings

This is the opposite of #98433 which concerned IDNA decoding.

IDNA uses the quadratic-complexity codec punycode, making it slow with large inputs.
We do check the limit (64 bytes per label), but too late (after the quadratic step is done).

Note that #155475 added documentation warnings, urging users to limit the input size:

• https://docs.python.org/3/library/codecs.html#text-encodings
• https://docs.python.org/3/library/codecs.html#module-encodings.idna

Linked PRs

• gh-157682