0.3.44

fix(merge-queue): re-run failed checks once before the first fix run

When CI failed, the queue started a fix run at once. On a flake there is
nothing to fix: on PostHog/posthog#125410 a runner timeout cost two fix
runs, each agent said a maintainer must re-run the job, and the entry
stopped with "no progress" and no rerun tried.

Now, when failing checks are the only blocker, the queue waits for the
rest of CI, re-runs the failed checks once, and starts a fix run only if
they fail again. If the rerun is impossible (for example, the App has no
Actions write permission), the fix run starts in the same pass and the
entry is not blocked. A conflict, requested changes, a branch behind its
base or unsigned commits still start the fix run at once. This rerun does
not use the update-branch fallback, because that push resets the budget.