Escape U+0085, U+2028 and U+2029 in NDJSON records

[sidkmenon]

JSON allows U+2028 (LINE SEPARATOR) and U+2029 (PARAGRAPH SEPARATOR) raw inside strings, so diffr wrote them raw when a diffed file held one. Since Node 24, node:readline also ends a line at these characters, so a client that reads diffr with readline gets a record cut in two and fails to parse it. Whiteboard's structural diff stops with Malformed diffr protocol record. (devdotfast/whiteboard#624).

Python's str.splitlines() has the same fault for U+0085 (NEL). Every other character it splits on is a control character that JSON already escapes.

• Adds protocol::write_record. Its formatter writes the same compact JSON, and escapes the three characters as JSON \uXXXX escapes. The decoded text and its byte columns do not change.
• run.rs writes every record through it.
• The protocol docs now say that records end at \n only and that the three characters are escaped.

Testing

• New unit test: the three characters are escaped, and the record parses back to the same event.
• The issue's repro on Node 24.18: the output has no raw U+2028, and readline reads 3 records that all parse.
• A file that holds all nine characters readline or splitlines() end a line at: splitlines() reads 3 records that all parse.
• cargo test, cargo clippy --all-targets (no new warnings) and cargo fmt --all -- --check passed locally.

AI assistance: written with Claude Code; reviewed by the author.