Harden Codex readiness beyond TUI copy and footer heuristics

[ynadge]

Context

PR #700 fixes initial-prompt delivery for Codex CLI 0.143 and adds a best-effort copy-independent composer proof. The current fallback recognizes known empty-composer copy or the › input marker paired with Codex's cwd footer, while any known onboarding fragment vetoes readiness.

Codex still does not create rollout/session metadata until its first turn, so Agent Studio needs a pre-turn readiness signal that cannot rely on that metadata.

Remaining limitation

After a trust/sign-in/setup screen has been observed, the safety latch intentionally has no time-only escape. If a future Codex TUI changes both its placeholder copy and its cwd footer shape (or hides the footer at a narrow terminal width), Agent Studio can safely remain unready forever instead of risking injection into a modal.

Goal

Replace or supplement TUI-copy/footer heuristics with a stable, version-independent readiness proof. Investigate, in order:

1. a structured lifecycle/readiness signal from Codex or Codex App Server before turn one;
2. a terminal-screen model that identifies the active input widget and excludes modal selection rows;
3. a bounded fallback that can prove multiple consecutive clean frames without weakening trust/sign-in safety.

Acceptance criteria

• New and resumed Codex sessions become ready across supported CLI versions and terminal widths without enumerating placeholder copy.
• Trust, sign-in, migration, personality, and theme modals never receive injected prompt text.
• Real PTY fixtures cover full and partial Ratatui repaints, including an underlying footer retained beneath a modal.
• Claude Code readiness behavior remains unchanged.