tui: bind Ctrl+Z (byte 0x1A) to the suspend handshake + add the cucumber PTY test (#6169 follow-ups)

Two follow-ups from the #6169 landing (`95bc5c008`), both deliberately deferred there:

1. **Ctrl+Z binding.** Raw mode means the keyboard Ctrl+Z arrives as byte `0x1A`, never SIGTSTP (`lib.rs:2145-2150` documents the same for Ctrl+C/0x03). The signal handshake now covers externally delivered SIGTSTP/SIGTTIN only; the vim-like Ctrl+Z needs a key handler that runs the full crossterm teardown (legal in normal context) and ends with `libc::raise(libc::SIGSTOP)` as the last statement. Entry points: composer/key routing in `crates/tui/src/tui/ui/handlers/`, `pause_terminal(...)` (`terminal.rs:432`).

2. **Cucumber PTY test.** Add `signal()` / `wait_until_stopped()` to `crates/tui/tests/support/qa_harness/pty.rs` and a `job_control_suspend_pty.rs` precedent per `screen_mode_inline_pty.rs`: deliver SIGTSTP, assert WUNTRACED stop + restore bytes on the control stream, SIGCONT -> repaint. The standalone smoke `artifacts/agent-reports/plans/6169-pty-smoke.py` proves the shape today; this asks for it in-repo under CI.

Also noted from the #6169 review: the background-SIGTTIN path is covered structurally (one shared handler + SIGTTOU ignored at install) but was not driven live end-to-end — the cucumber test above is where that gets proven.

Read the composer/key routing handlers under crates/tui/src/tui/ui/handlers/ and pause_terminal(...) in terminal.rs:432, then inspect the Ctrl+C/0x03 notes at lib.rs:2145-2150. Add the Ctrl+Z suspend path and extend crates/tui/tests/support/qa_harness/pty.rs with signal() and wait_until_stopped(); use screen_mode_inline_pty.rs and the smoke script as precedents. Done means the PTY test verifies stop, terminal restoration, SIGCONT, and repaint.

Written by the indexing model from the issue text.