kernel/tests: de-flake the Wave-1b input-back integration test

The surface-element tree + input-back case drove a real xdg client from a free-
running thread; under load that thread was starved, so its async receive
assertions (ptr/touch/keyboard enter + coords) either read sentinel values and
FAILED or the whole body silently SKIPPED -- the suite's assertion count
oscillated (375 skip / 407 race) and the kernel gate was unreliable.

Drive the client deterministically with condition-based cooperative event-loop
pumping (no time-based waits, no skip), so the input-back / tree / transformed-pick
/ keyboard / press-hook coverage runs and passes every time. No production change.

Verified: 8 consecutive normal runs + asan all OK with an identical doctest
assertion count (391, 0 failed) every run.

0 files changed, 0 insertions, 0 deletions