Fix another overflow in punycode encode_into

[Skgland]

It was possible to panic due to an overflow in build with panic on overflow (i.e. debug builds), if the input iterator contained more than u32::MAX elements.
Which can happen an systems with target_pointer_width > 32 if a hugh string or char slice is passed to encode or encode_str respectively.

Commit 1. adds a failing test
Commit 2. fixes the implementation so that the test passes
Commit 3. adds some early detection to encode and encode_str to not waisted work allocating the String and iterating the input

[codecov]

Codecov Report

Attention: 3 lines in your changes are missing coverage. Please review.

Comparison is base (912d716) 81.85% compared to head (9b16473) 81.75%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #880      +/-   ##
==========================================
- Coverage   81.85%   81.75%   -0.10%     
==========================================
  Files          20       20              
  Lines        3532     3536       +4     
==========================================
  Hits         2891     2891              
- Misses        641      645       +4     

Files	Coverage Δ
idna/src/punycode.rs	83.95% <50.00%> (-1.50%)	⬇️

... and 1 file with indirect coverage changes

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[valenting]

Thanks for the fix.
I'll land this as is. I think we actually want to limit the input size even more, as IDNA labels that exceed 63 characters are not allowed in DNS anyway.