DEN-4233: preserve UTF-8 across JSON stream chunk boundaries

[ORESoftware]

Why

The Transform decoded each incoming Buffer independently with String(chunk). A valid UTF-8 code point split across stream chunks could therefore be replaced/corrupted before JSON parsing, including when a multibyte custom delimiter itself crossed a byte boundary.

Changes

• decode byte streams incrementally with Node StringDecoder('utf8');
• preserve accurate byte-read accounting from the actual byte buffer;
• flush decoder tail before parsing the final JSON fragment;
• test every possible byte split for JSON containing café, emoji, CJK text, and a multibyte ∆∆∆ delimiter;
• add a pinned Node 22 GitHub Actions regression lane using the existing JS/TS parser tests without requiring the optional native addon build.

Contract note

This parser is an in-process byte-to-object library, so this PR intentionally does not invent a TypeSpec/JSON Schema authority. TJSV remains required where a real serialized cross-runtime contract exists; the gRPC reconciliation repo is the audited boundary target for that enforcement.

Linear: DEN-4233

[linear-code]

DEN-4233