test(cli): await attached input readiness in held-frame regression - #1129

Why

Main CI fails in the attached first-frame regression introduced by #1124. Reaching the held first content write does not prove that the separate terminal input reader has subscribed; Bun 1.4.0 observed zero readers in both CI attempts.

Closes #1128. Repairs the test failure tracked by #1126.

What changes

Wait for the actual input subscription through the existing bounded awaited helper before asserting the exact reader count and feeding input. The first content write stays held. Every original assertion remains: blank screen, one reader, detach without another key, zero prompts, one reset, no remaining readers/listeners and subsequent child usability.

Only packages/cli/tests/repl-headless-attach.test.ts changes (+4/-1). Product behavior, timeouts, runtime selection, dependencies and generated files are unchanged.

Validation

Signed feedback commit 25576418d2ffdd6d2908273aec6cf00528de4019, based on merged main 80959039dc0fbca2cc4c047941a5e23cc6ed5178:

• Focused Bun file: 12 passed, 0 failed.
• Focused Deno file: 5 passed, 12 steps, 0 failed.
• A controlled delayed reader subscription makes the original assertion fail and the corrected test pass under both local runtimes, with the content write still held.
• Original Planner and the same independent Astra reviewer both passed this exact commit with no findings. Required CI is pending; the maintainer decision on the main-repair label is pending.

Local Bun is 1.3.14, while the CI failure is on Bun 1.4.0. Local baseline passed; the controlled schedule reproduces the missing precondition without claiming a natural local 1.4.0 reproduction. No benchmark repeats for a test-only change.

Scope confirmation

• One test correction using an existing bounded helper.
• All existing behavioral assertions remain effective.
• No product or dependency changes.
• The patch release remains a separate delivery after this repair.