fix(tests): make the resize waits one-directional

[vyncint]

The stress gate caught this on 2.0.0's own change — hunt (ubuntu-latest, 16 threads), iteration 10 of 25. Which is what the gate is for.

resize_relayouts_the_frame ... FAILED
the relaid-out frame: Timeout { waiting_for: "a complete frame matching the predicate —
the application has not completed a repaint since the frame this terminal last returned
(2 complete frames in total)..." }

Cause

resize_relayouts_the_frame waited on ready before resizing, and ready is true of every frame. wait_frame returns the earliest frame nobody has observed, so under load that first call could hand back the frame the resize had already produced — leaving no later frame for the second wait.

Fix

Both waits are one-directional now. 2.0.0 supplied the discriminator for free: the chosen line has room for its confidence interval at 110 columns and not at 80, so the interval's absence identifies a pre-resize frame and its presence a post-resize one. Neither predicate can be satisfied by a frame from the other side of the resize.

Audited the rest

Every other follow-up predicate is already one-directional — ranking (, all refused configurations:, measured 11 of, and the scrolled-row check are each true only after their keypress — and the two bare wait_frame(ready) calls are the only wait on a freshly spawned terminal. Resize was the sole place two predicates could match the same frame class.

Verification

25 local iterations at 16 threads, no flake — which is not proof, since the CI failure was at iteration 10. The stress matrix is, and I will run it before tagging.

Checklist

• All commits signed off
• No AI attribution trailers
• No golden changes — rendering is untouched