test(web): H2 real transcript corpus + parser-fidelity tests (WI-123)

[thedancingdeveloper]

Fifth PR of the terminal-attach-budget initiative (WI-121). Stacked on #670 (F2); all land on feat/terminal-attach-budget.

What this adds (H2 / WI-123)

The ground-state trimming both the server ring and the client cache rely on was only tested against synthetic line\n data. This adds a corpus of real PTY captures and a fidelity suite over it.

• web/tests/fixtures/transcripts/ (gzipped, sanitised):
  • claude-code-tui — a curses alt-screen redraw stream (cursor addressing, SGR, OSC title, box/emoji UTF-8). It emits no line feeds — the seamless case where raw-byte trimming has nothing to align to.
  • shell-plain — coloured recursive ls + multibyte box/CJK/emoji chars.
  • cargo-build — real cargo build -v colour output with \r.
  • README.md documents the intended production source (GET /api/history/:id/download from a dev stack) and how these local stand-ins were captured.
• scripts/capture_transcript.py — a pty.fork capture helper.
• scripts/sanitise_transcript.py — rewrites home paths / tokens / JWTs / emails to inert placeholders and asserts no secret pattern survives before a fixture is committed.
• src/__tests__/transcriptFidelity.test.ts — over each corpus at 4 KiB cut offsets, groundStateReplayStart / prepareReplayTail / sliceForReplay pick a start never inside an escape or on a UTF-8 continuation byte (just past a \n, or start-of-stream), and the kept tail is byte-for-byte the source suffix.

Scope note

The plan's xterm serialize()-compare fidelity and the parse-rate baseline both need @xterm/addon-serialize. Rather than add a test-only copy here, they fold into F5 (WI-129), where that addon becomes a real app dependency. This PR delivers the corpus, sanitiser, and the ground-state fidelity that F5 and H3 build on.

The alt-screen corpus makes a real limitation explicit: a newline-free stream has no seam the raw-byte trim can prove is ground state, so it falls back to replaying as-is — exactly what F5's serialized-state restore closes.

Verification

• pnpm vitest — 928 passed (15 new).
• ruff check + ruff format --check clean on the scripts; scripts/check_docs.py clean.

🤖 Generated with Claude Code

https://claude.ai/code/session_01UtEFLJAhLiq4NfZCN2vuBb

[thedancingdeveloper]

Cleanup: superseded by the single promotion PR #676. All of this work is already integrated into feat/terminal-attach-budget and is reviewable there per-commit (each commit is one initiative item with a full message). Closing to keep the initiative to one PR; the commit is preserved.