fix: use StringDecoder to handle UTF-8 chunk boundaries in setEncoding

[398651434]

Description

Fixes a bug where response.body.setEncoding('utf8') corrupts multi-byte UTF-8 characters that span chunk boundaries.

Root Cause

Each chunk was being individually converted to a string via buffer.utf8Slice() (or toString()). When a multi-byte UTF-8 character (e.g., a Chinese character = 3 bytes) is split across two HTTP response chunks, the first chunk gets an incomplete byte sequence converted to garbage, and the second chunk's portion becomes a separate corrupted character.

Fix

Use Node.js's built-in StringDecoder (from node:string_decoder) which properly buffers incomplete byte sequences between write() calls:

1. setEncoding(encoding): Initialize a StringDecoder when encoding is set
2. consumePush: When a decoder exists, use decoder.write(chunk) instead of storing the raw buffer — this accumulates incomplete UTF-8 bytes internally
3. consumeFinish: Reset the decoder to allow garbage collection

Testing

The bug manifests when:

• HTTP response contains multi-byte UTF-8 text (e.g., Chinese characters, emoji)
• setEncoding('utf8') is called on the body
• The text spans multiple TCP packets/chunks

After fix, characters are correctly reassembled across chunk boundaries.

---

Closes #5002