fix: fork_session drops transcript entries containing U+2028/U+2029/U+0085 - #1495

Problem

_parse_fork_transcript split the transcript with str.splitlines(), which also breaks on U+2028 (LINE SEPARATOR), U+2029 (PARAGRAPH SEPARATOR) and U+0085 (NEL) — characters Claude Code's JS JSON.stringify leaves unescaped inside strings (valid JSON). A single record holding one of them was cut into pieces that all failed json.loads and were silently skipped by the except: continue branch. The entry disappeared from the fork and the next entry's parentUuid resolved to None, breaking the resumed chain — resuming the fork then loaded only the part after the break.

Fix

Split on "\n" only — the record separator the CLI actually writes — the same approach _parse_transcript_entries in sessions.py already uses. Behavior for well-formed JSONL is unchanged; a comment documents why splitlines() must not be reintroduced here.

Testing

• New test_fork_preserves_entries_with_unicode_line_separators writes a 3-entry transcript (the middle one containing all three separator characters, unescaped as Claude Code writes it), forks it, and asserts all three messages survive with an intact parent chain. Red before the fix (fork contained ['first', 'third']), green after.
• Full suite passes; ruff check, ruff format, mypy clean.

Fixes #1441