compose: Enforce max topic/content length by Unicode code points, not UTF-16 code units

[chrisbobbe]

We have this comment where we decide if the content input is too long:

      // normalized.length is the number of UTF-16 code units, while the server
      // API expresses the max in Unicode code points. So this comparison will
      // be conservative and may cut the user off shorter than necessary.
      if (textNormalized.length > kMaxMessageLengthCodePoints)

This issue is to stop cutting the user off shorter than necessary, while not regressing on performance in the common case where the length isn't approaching the limit.

This applies equally to topic lengths, where the server limit is also expressed in Unicode code points. For example, on CZO where the topic limit is 60 code points, we're not accepting '👨‍👩‍👦👨‍👩‍👦👨‍👩‍👦👨‍👩‍👦👨‍👩‍👦👨‍👩‍👦👨‍👩‍👦👨‍👩‍👦👨‍👩‍👦👨‍👩‍👦👨‍👩‍👦👨‍👩‍👦' (60 code points), but we should accept it because the server does.