Disk-gated build queues: re-queue skipped jobs with a free-space waiter

When a shared local build queue skips jobs below a free-disk floor, the skip often happens because the previous build's temporary files are still on disk, not because the disk is truly full. Free space usually comes back 1-2 GB a minute or so after the previous job ends. Instead of re-queuing by hand, run a background waiter: until [ $(df -g / | awk 'NR==2{print $4}') -ge FLOOR ]; do sleep 30; done; echo '<sha> <label>' >> queue. Have it print the free space when it fires so the log shows why the job resumed. Also ask the agents that just finished to delete their regenerable build dirs right away; that keeps the next job from being skipped.