ci: GitHub Actions stalled ~6h — no run since 13:08Z, no queued (self-hosted runners offline); blocks nightly + releases + PR CI

[avrabe]

Observed (gale, measured): the most recent gale Actions run started 2026-06-14T13:08:34Z (the PR #61 merge batch). As of ~19:07Z — ~6 h — there has been no new run and nothing queued (gh run list max createdAt = 13:08:34Z; zero status!=completed). So it's not a backlog behind busy runners — runs aren't being created/dispatched at all (self-hosted smithy runners appear offline; flagged independently on loom#219).

Correlated symptom: a local bazel test //:verus_test hangs in analysis on the rocq_nixpkgs nix-build (network fetch stuck) — same network/infra trouble window.

Impact (what's blocked until runners are back):

• The nightly (mps2/an521/cpu0 board-target oracle from fix(nightly): cortex-m33 board target mps2/an521 → mps2/an521/cpu0 (4.4.0 dual-cluster) #61) — can't validate.
• gale releases + the wasm-module release lane (release-wasm.yml).
• All PR CI — e.g. PR silicon: wasm-cross-LTO multi-target (ARM+RISC-V) — testbed, on-silicon results, smart-data #40 pushes since 13:08 triggered nothing.
• Downstream: the loom#219 SROA-fix and synth#354/#209 release-verification timing depend on CI recovering.

Not a code regression — PR #61 (the 13:08 batch) ran and completed; the stall is after 13:08 (infra). Action needed: bring the self-hosted runners back online (ops), then re-trigger the nightly + PR CI. Filing so the outage is tracked rather than silently swallowing the nightly + release gates.

Kill-criterion (resolved): a gale Actions run with createdAt > 13:08:34Z completes (runners dispatching again).