TUI ignores suppressOutput for prompt hook context

What version of Codex CLI is running?

codex-cli 0.149.1; the behavior is also present on main at 9dea1f709f.

What platform is your computer?

Microsoft Windows NT 10.0.26200.0 x64, Codex TUI.

What issue are you seeing?

A successful UserPromptSubmit hook that returns model-facing additionalContext
is rendered in the TUI even when the structured output sets suppressOutput to
true:

UserPromptSubmit hook (completed)
  hook context: commit-commands runtime metadata for this turn: ...

The hook output parser preserves suppressOutput, but the session-start and
user-prompt-submit event parsers currently discard it. The context therefore
becomes both model input and a visible HookOutputEntryKind::Context entry.

What steps can reproduce the bug?

1. Configure a synchronous UserPromptSubmit command hook.
2. Have the hook print this structured output:

{
  "suppressOutput": true,
  "hookSpecificOutput": {
    "hookEventName": "UserPromptSubmit",
    "additionalContext": "quiet model context"
  }
}

3. Submit a prompt in the Codex TUI.
4. Observe that the successful hook and its context are displayed.

The same parsing issue applies to SessionStart and SubagentStart context.

What is the expected behavior?

The additional context should still reach the model, but a successful hook
with suppressOutput: true should not produce a visible context entry in the
TUI. Block, stop, warning, and error feedback should remain visible.

Additional information

The TUI already omits completed hook runs whose output entry list is empty, so
the event parser can implement this without a protocol change: keep the
model-facing context while omitting only the visible context entry.

Start with the session-start and user-prompt-submit event parsers, then trace how the hook output parser preserves suppressOutput and how the TUI builds HookOutputEntryKind::Context entries. Keep additionalContext in model input while omitting only the visible context entry when suppressOutput is true; block, stop, warning, and error feedback should remain visible.