Summary

• Test suites pull many testcontainers images over the course of a single CI job (elasticsearch, kafka connect, etc.), which can exhaust runner disk space mid-run — saw this cause a spurious CI failure on Consolidate markdown formatting onto rumdl/flint, drop spotless prettier #19183 (docs-only PR, No space left on device on a test3 matrix job, disk went from 39GB free to 0 during the test run).
• Adds docker system prune -af --volumes to .github/scripts/gha-free-disk-space.sh, removing unused Docker content already present on the runner before tests begin.
• GitHub-hosted jobs start on a fresh runner VM, so this is not intended to clean Docker data from a prior workflow run. It targets unused content present on the fresh runner image or created during job setup.

Caveat: this only frees space before tests start, matching the existing script's approach — it does not address images pulled during the job, so it may reduce but not eliminate this failure mode.

Observed evidence

• The 16 Ubuntu smoke jobs in run 29429470350 and run 29580716222 each reported 1.851 GB reclaimed by Docker pruning.
• The overall filesystem cleanup increased available space by roughly 25–26 GB in those jobs; that larger number includes the script's non-Docker cleanup and should not be attributed to Docker alone.
• Both runs completed successfully. This is evidence of additional capacity, not a controlled measurement of failure prevention or total runtime improvement.

Test plan

• CI passes