Make the REPL Tab focus test finish consistently #1015

The REPL Tab focus test finishes consistently while checking that Tab moves the visible cue and keyboard focus in one presented frame.

During the behavioral test cleanup in #1013, RS1: Tab moves the cue and the focus, and presents once intermittently timed out after 15 seconds. The saved native command result shows a passing full-file run followed by a failing run in the same two-run command, with no source-writing command between those invocations.

Evidence

• Local Deno 2.9.7 on aarch64 Apple Darwin; no CI run link exists for this observation.
• Command: deno test -A --frozen packages/cli/tests/repl-journey.test.ts.
• Checkout HEAD was 541fd0df, with the journey migration still uncommitted. The exact transient worktree hash was not saved.
• The native tool result at 2026-10-29 19:03:43 UTC preserves this output:

ok | 42 passed (77 steps) | 0 failed (26s)
  RS1: Tab moves the cue and the focus, and presents once ... FAILED (15s)
REPL responsiveness: a focus move is one frame ... FAILED (due to 1 failed step) (15s)
FAILED | 41 passed (76 steps) | 1 failed (1 step) (42s)

The complete timeout stack was not preserved. The migration helper subsequently added a wait for the draft to offer Send before starting the focus measurement and reported passing later runs. A late startup frame is a hypothesis, not an established cause from the available failure evidence.

Acceptance

Preserve the existing cue, keyboard-focus and single-presentation assertions. Identify the scheduling cause and record a regression that distinguishes the repair from the failing behavior. Do not replace those assertions with a delay or an ignored failure.

This is the dedicated flake record required by AGENTS.md. Close it only after the cause, regression evidence and delivery are recorded; a green rerun alone does not close it.