Expose alternate-screen and slave input modes in PTY status

[myobie]

Problem

A live PTY can remain in a Vim/Codex-like full-screen and raw-input state after detach. PTY already tracks DEC alternate-screen entry/exit internally (altScreenActive) plus mouse, cursor, and Kitty keyboard modes, but live stats omit alternate-screen. PTY does not expose the slave TTY termios state, so consumers cannot distinguish canonical input from noncanonical/raw input through a stable API.

TERMINAL_SANITIZE restores the attaching operator's host terminal on detach; it does not change the child PTY's alternate buffer or slave line discipline.

Read-only contract

Extend existing live status diagnostics without making them an idle oracle:

modes: {
  alternateScreen: boolean;
  bracketedPaste: boolean;
  applicationCursor: boolean;
  sgrMouse: boolean;
  cursorHidden: boolean;
  kittyKeyboard: boolean;
  kittyKeyboardFlags: number[];
  input: {
    state: "canonical" | "noncanonical" | "raw" | "unknown";
    canonical: boolean | null;
    echo: boolean | null;
    signals: boolean | null;
    extended: boolean | null;
    min?: number;
    time?: number;
  };
}

• alternateScreen comes from the daemon's existing DECSET/DECRST tracking, cross-checked with xterm's active buffer.
• Exact Unix input state must come from tcgetattr on the PTY slave/master through a supported node-pty/native boundary; do not parse screen text or infer raw mode from Kitty/mouse/cursor sequences.
• Unsupported platforms return unknown, never a guessed safe value.
• These are diagnostics/policy inputs, not semantic idle. Codex can legitimately be idle at its prompt while alternate-screen and raw input remain active; a generic unknown-harness policy may still choose to withhold typing.

Current observed proof

The live Codex PTY remained process-alive with no attached viewers while its slave TTY reported -icanon -isig -iexten -echo, min=1, time=0, raw-style input/output flags, and CS8. Live PTY stats simultaneously reported Kitty keyboard protocol active. That proves the child PTY stays in a raw application-input mode; current public stats cannot prove alternate-screen because the tracked flag is omitted.

Model-free acceptance tests

Use stable ASCII screen plus termios/mode fixtures for:

• alternate-screen enter and retained state after detach;
• raw/noncanonical slave mode and long-running child;
• terminal restore (DECRST plus canonical termios);
• clear/compaction without false restoration;
• child crash while modes remain diagnosable until daemon cleanup;
• daemon restart resets output-mode trackers and reports input unknown until observed;
• stale screen/event logs cannot populate live mode state.

No deployment. Keep harness-specific idle decisions outside PTY core.