CI: self-hosted runner9 disk-full — Code Coverage failing with 'No space left on device' (will hit required jobs next)

[avrabe]

The Code Coverage job is failing on runner9 with System.IO.IOException: No space left on device (seen on main HEAD fd18d56 and PR #571, 2026-07-02). Today it only reddens the advisory codecov job, but the same runner serves required jobs — this becomes a hard CI outage when a required job lands on it.

Known recurring culprits on runner disks (from prior local episodes): target/debug/incremental ballooning to ~35 GB, stale .claude/worktrees agent dirs, old cargo caches. Suggested remediation on runner9 (whoever owns the box):

• cargo sweep / delete */target/debug/incremental older than a few days
• prune stale worktrees + docker system prune if applicable
• consider a scheduled cleanup step (pre-job df guard + cache eviction) in the runner's maintenance, so this stops recurring

Filed by the release-gate loop; not blocking current merges (codecov is advisory) but tracked so it's fixed before it blocks a real gate.

[avrabe]

Spreading: runner7 now too. Fuzz Smoke wasm_ops_lower_or_error on main c766c2a (v0.22.0) died with System.IO.IOException: No space left on device: '/var/lib/runners/runner7/_diag/...' (run 28585412930) — the job failed writing its own diagnostics log, not on any code path (the release commit is version-pins-only; the other 3 fuzz jobs passed). Two runners (runner7, runner9) are now disk-full. This needs the fleet-wide cleanup, not per-runner triage — and it's begun producing false-negative signal on main (a 'fuzz failure' that's pure infra).