fix(ci): purge corrupted buildx cache on crc32/unpigz failures

Buildx failed with unpigz crc32 mismatch; retry now does full prune, drops stale builder volumes, and uses --no-cache on the last attempt.

scripts/retry-docker-build.sh

@@ -1,5 +1,5 @@
 #!/usr/bin/env bash
-# Retry docker/buildx commands on transient network/registry failures.
+# Retry docker/buildx commands on transient network/registry/cache failures.
 # Usage: retry-docker-build.sh [max_attempts] [wait_seconds] -- <command...>
 set -euo pipefail
 
@@ -13,22 +13,59 @@ if [[ $# -eq 0 ]]; then
   exit 2
 fi
 
+is_corrupt_cache() {
+  grep -qiE 'crc32 mismatch|unpigz|corrupted|failed to compute cache key' <<<"$1"
+}
+
+purge_buildx_cache() {
+  echo "== purge buildx cache (all + stale builders)"
+  docker buildx prune -af >/dev/null 2>&1 || true
+  docker builder prune -af >/dev/null 2>&1 || true
+  docker ps -aq --filter "name=buildx_buildkit" --filter "status=exited" |
+    xargs -r docker rm -f >/dev/null 2>&1 || true
+  docker volume ls -q | grep -E '^buildx_buildkit_.*_state$' | while read -r vol; do
+    if docker ps --filter "volume=${vol}" --format '{{.ID}}' | grep -q .; then
+      continue
+    fi
+    echo "== remove volume ${vol}"
+    docker volume rm -f "${vol}" >/dev/null 2>&1 || true
+  done
+}
+
 attempt=1
+corrupt_seen=0
 while true; do
-  echo "== docker build attempt ${attempt}/${MAX}: $*"
+  cmd=("$@")
+  # After a corrupt-cache failure, force a clean rebuild on the last attempt.
+  if (( corrupt_seen == 1 && attempt == MAX )); then
+    if [[ "${cmd[*]}" == *buildx\ build* ]] && [[ "${cmd[*]}" != *--no-cache* ]]; then
+      echo "== last attempt: injecting --no-cache"
+      cmd+=(--no-cache)
+    fi
+  fi
+  echo "== docker build attempt ${attempt}/${MAX}: ${cmd[*]}"
+  logf="$(mktemp)"
   set +e
-  "$@"
+  "${cmd[@]}" 2>&1 | tee "${logf}"
-  code=$?
+  code=${PIPESTATUS[0]}
   set -e
   if (( code == 0 )); then
+    rm -f "${logf}"
     exit 0
   fi
+  log="$(cat "${logf}" 2>/dev/null || true)"
+  rm -f "${logf}"
   if (( attempt >= MAX )); then
     echo "== all ${MAX} attempts failed (last exit ${code})"
     exit "${code}"
   fi
   echo "== attempt ${attempt} failed (exit ${code}), retry in ${WAIT}s..."
-  docker buildx prune -f >/dev/null 2>&1 || true
+  if is_corrupt_cache "${log}"; then
+    corrupt_seen=1
+    purge_buildx_cache
+  else
+    docker buildx prune -f >/dev/null 2>&1 || true
+  fi
   sleep "${WAIT}"
   attempt=$((attempt + 1))
 done