feat(terminal): replay 10,000 retained rows without cell-grid copies

Serialize the worker-owned terminal directly so larger retained history does not require a second object grid. Commit input modes with parsed output so replay bytes and metadata describe the same revision.

docs/terminal-performance.md

@@ -19,13 +19,14 @@ Terminal frames share the daemon main event loop with all agent traffic. The `ev
 
 ## Invariants (the easy-to-break ones)
 
-- **Replay is incremental across the daemon and SDK.** The worker retains 1,000 history rows. Capable clients request their configured history; the worker bounds it by retained history. Older apps retain their own 200-row request. Explicit SDK history demand is unchanged against older ANSI hosts, which apply their existing 500-row ceiling. Begin resets once, chunks keep a streaming UTF-8 decoder/parser, and each renderer suppresses generated replies while parsing replay. Never concatenate capable replay chunks to feed the renderer.
+- **Replay is incremental across the daemon and SDK.** Capable clients request their configured history, bounded by [worker retention](terminal-worker.md#process-model). Older apps retain their own 200-row request. Explicit SDK history demand is unchanged against older ANSI hosts, which apply their existing 500-row ceiling. Begin resets once, chunks keep a streaming UTF-8 decoder/parser, and each renderer suppresses generated replies while parsing replay. Never concatenate capable replay chunks to feed the renderer.
+- **Replay serialization reads xterm directly.** The [worker](terminal-worker.md#process-model) uses its existing buffer; text capture translates only the requested rows, and readiness scans row strings. Do not rebuild a full cell-object grid. Serialization still allocates a whole ANSI string and UTF-8 bytes before bounded delivery; chunking does not bound those allocations.
 - **Coalescers are leading+trailing throttles.** The first chunk after an idle window flushes immediately (synchronously); only sustained bursts wait for the trailing timer. Reverting to trailing-only adds a full window (~5ms) to every keystroke echo.
-- **Output coalescing happens in the worker, before IPC.** One socket frame per pty chunk was a main-loop flood under build output. Non-output messages (snapshot/snapshotReady/titleChange/exit) must flush the coalescer first so ordering is preserved.
+- **Output coalescing happens in the worker, before IPC.** One socket frame per pty chunk was a main-loop flood under build output. Non-output messages (snapshotReady/titleChange/exit) must flush the coalescer first so ordering is preserved.
 - **Replay delivery is isolated per viewer.** Live worker output never waits for app delivery. Each daemon stream owns its bounded live tail and revision deduplication; a slow or failed replay cannot pause another viewer of the same terminal.
 - **Replay geometry belongs to the parser queue.** Native measurements and web fits must leave the captured dimensions intact through every chunk's commit. Apply the latest measured size after replay, before live bytes; clear, replacement and disposal must release that geometry ownership. A layout effect must not resize the renderer directly during replay.
 - **Coalesced output carries the LAST chunk's revision.** Snapshot replay dedup (`replayTerminalOutputAfterSnapshot`) skips buffered output with `revision <= replayRevision`; a merged batch with a lower revision would be wrongly skipped (lost output).
-- **The input-mode tracker runs once per process boundary, not per hop.** The worker owns the authoritative tracker; the daemon caches current modes from ordered worker events and adoption. A replay response carries capture-time modes for its viewer; it must not overwrite the shared cache after newer live output. Apply that viewer's preamble before its buffered live bytes. Do not reintroduce a per-chunk `feed()` on the daemon main loop.
+- **The input-mode tracker runs once per process boundary, not per hop.** The worker commits its authoritative tracker with xterm parsing before publishing the corresponding revision; the daemon caches current modes from ordered worker events and adoption. A replay response carries capture-time modes for its viewer; it must not overwrite the shared cache after newer live output. Apply that viewer's preamble before its buffered live bytes. Do not reintroduce a per-chunk `feed()` on the daemon main loop.
 - **Snapshot catch-up is backpressure-gated.** A stream falls back to a full snapshot only when `outputBytesSinceSnapshot > MAX_TERMINAL_OUTPUT_FRAME_BYTES` (256KB) **and** the client transport reports `bufferedAmount > MAX_CLIENT_BUFFERED_BYTES` (4MB). A client that keeps draining streams continuously, no matter how much output is produced. Before this gate existed, every 256KB of build output dropped a frame and forced a full JSON cell-grid snapshot (~200k objects across IPC) — the historical source of spiky lag and GC hitches.
 - **Plugin daemon sessions report IPC queue bytes.** Their virtual socket increments `bufferedAmount` before `process.send` and decrements it only in the send callback. Text and binary frames share that ordered queue, so the normal snapshot catch-up and physical high-water gates remain valid for server-side plugin SDK traffic.
 - **Client output writes are not serialized per frame.** The emulator runtime drains contiguous plain writes straight into xterm (which buffers internally). Only barrier ops (`clear`, `snapshot`, `suppressInput` writes) wait — behind a zero-length sentinel write — so resets can't interleave with in-flight output.

docs/terminal-worker.md

@@ -15,6 +15,8 @@ supervisor (paseo.pid) ──► daemon worker ──► launcher ──► term
 - The worker exits on its own when it has no controller and no terminals, on `shutdown`, on SIGTERM (killing its terminals), and when its record disappears (a deleted `$PASEO_HOME`: nobody can ever reach it again).
 - Its stderr goes to `$PASEO_HOME/terminal-worker.log`. No pipe to the daemon survives the daemon.
 
+The worker retains a fixed 10,000 history rows plus the screen in its existing xterm. Replay uses `@xterm/addon-serialize` 0.14 against headless xterm 6; no second cell grid or lifetime byte log is retained. This addon omits cursor visibility and DECSCUSR presentation, so the session supplements those sequences from xterm’s cursor state. Standard modes come from the serializer; the existing input-mode tracker retains protocol modes such as Kitty and Win32 input. Commit that tracker with parsed output so replay bytes and metadata describe the same cut.
+
 Which triggers keep terminals and which close them is owned by the stop reason (`keepTerminals`) that flows from the request through the supervisor to `stop()`. The kill side always ends in the reaper (`terminal-worker-reaper.ts`), the only path that signals terminal processes outside a connected daemon.
 
 ## Record, socket and secret