Docker build fails inode exhausted

A self-hosted CI runner host starts failing jobs at 09:00 with 'no space left on device' during docker build and git checkout, but df -h on /var/lib/docker shows only 40% bytes used. df -i shows that filesystem at 100% inodes. The host runs hundreds of short-lived containers a day; image pulls and intermediate build layers (each a tree of many small files in overlay2) have accumulated, and docker image prune/container cleanup has never been scheduled. No deploy, just steadily more pipelines over months. Triage and recover.