tmux forwards keystrokes to a pane's foreground process the instant
attach succeeds, whether or not that process has entered raw mode yet.
A full-screen TUI agent (Claude Code, Codex, ...) can take a moment
after launch to do so, so fast typing right after a session opens
lands as raw, echoed text in the scrollback instead of the agent's
input box.

Hold input on a pane's first-ever attach for a short grace window,
buffering keystrokes via the existing pendingInput mechanism instead
of forwarding them early. Only a pane's genuine first attach waits;
reattaching to an already-running pane or a second client is
unaffected.

Fixes Untrivial-ai#3023

Addresses review on the terminal input-hold fix (AgentWrapper#3023):

- The hold now starts on a pane's first SUCCESSFUL PTY publish instead
  of at open-request time. A shared one-shot inputGate per pane id
  replaces the eager deadline: only setPTY's own successful attach can
  arm it, so a slow Attach or a failed-then-retried attach can no
  longer let the window expire before the pane is even up.
- setPTY now publishes the Stream and returns immediately so copyOut
  starts reading output right away; releasing buffered input happens
  on a separate goroutine gated on the shared inputGate, so a boot-time
  output burst is no longer held hostage behind the input hold. The
  release goroutine checks the Stream is still current before flushing,
  so a stale waiter can't write into a Stream a later reattach replaced.
- The gate is kept for the Manager's lifetime instead of being
  time-pruned, so reopening a long-running pane can never be mistaken
  for a fresh one and re-delayed.

Adds regression tests for a slow attach, a failed-then-successful
attach, output streaming during the hold, and a reattach long after the
window elapsed.

…tion-safe

Addresses a second review round on the terminal input-hold fix (Untrivial-ai#3023):

- Reset the pane's shared input gate on every genuine (re)launch of its
  agent process (terminal.Manager.ResetInputGate), wired through a new
  session_manager.InputGateResetter hook called from relaunchSession.
  Without this, a resume/restart that reuses the runtime handle
  (ports.RuntimeRestarter) would inherit the exited process's
  already-open gate and let the replacement TUI's early keystrokes
  through unheld, reproducing the original race on every resume.
- releaseInput now re-reads the live Stream fresh before every single
  buffered write instead of holding one Stream reference for the whole
  drain, guarded by a draining flag so only one drain loop runs per
  attachment at a time. A Stream swap mid-drain (a reattach racing the
  release) now hands the remainder to whichever Stream is current,
  instead of risking a write into one this attachment no longer owns.
- inputGate.wait now reports whether the gate actually opened or ctx
  ended first; releaseInputWhenGateOpens abandons on the latter. Manager
  ../Close cancels its context before marking attachments closed, so a
  waiter could previously flush buffered keystrokes mid-shutdown.
- A genuine (still-current) write failure during the async flush now
  cancels the attach loop and closes the Stream, so run's blocked
  copyOut unblocks and the loop actually exits instead of treating a
  dead Stream as an ordinary drop-and-reattach case.
- releaseInput takes a context.Context (AGENTS.md: ctx first-arg for
  I/O/blocking work) and checks cancellation before draining each chunk.

Adds regression tests for: a resume resetting the gate, a Stream swap
mid-drain preserving order across the two Streams, Manager.Close
abandoning a mid-hold flush promptly, and an async flush failure
closing its Stream and ending the attach loop.