fix: preserve utf8 multibyte characters split across chunks in XmlTransformStream

[mattermoran]

Fixes an issue where multibyte utf8 characters like ö or emojis would get corrupted if their byte sequence was split across multiple stream chunks.

Previously the XmlTransformStream used textDecoder.decode(chunk) per chunk. If a multibyte char was split across chunks the decoder would treat an incomplete sequence as invalid and replace it with a replacement char U+FFFD (�)

The fix is to call textDecoder.decode(chunk, { stream: true }) which tells the decoder to buffer incomplete sequences between chunks.
The final textDecoder.decode() in flush emits any buffered sequences

Before: ö��ö
After: ööö

links:

• WHATWG Encoding spec
• MDN TextDecoder