It'd be nice to have convert_utf16_to_utf8_{with_errors}_safe that takes an output size and only writes as much as that size. I noticed that there's a output size-taking variant for Latin1, convert_latin1_to_utf8_safe, so there's precedent for this. Understandably this will be slower than assuming the output is sufficiently large in that there'll be a bunch of additional bounds checks.

The full context is that related to #688, another missing piece to using simdutf in V8 is that V8's String::WriteUtf8V2 API tries to be safe and takes an output size. If the required capacity exceeds the actual capacity during transcoding, we try to write as much as can fit into the actual capacity.

That is, while V8's API is also intended to be used in a 2-phase mode of first computing the needed length via String::Utf8LengthV2, there's extra flexibility into allowing people to do partial transcoding bounded by output size. simdutf only allows partial transcoding bounded by input size.