Use correct byte size for truncation

[waltjones]

When calculating payload length for truncation, payload.length was being used. This returns the character count (or more accurately the UTF-16 code point count), rather than the UTF-8 byte count. This leads to non-ascii payloads being insufficiently truncated and then rejected by the API.

This PR uses a minimal implementation for counting UTF-8 bytes that never undercounts bytes, and rarely overcounts*. This is safe and valid to be used for truncation, and is significantly smaller and faster than a complete UTF-8 encoder.

• It will overcount when it encounters UTF-16 surrogates (two code point chars.) These should be rare in practice, but specific support for these could be added if/when desired.