ci(deploy): retry the :22 smoke test for up to ~2 minutes (#137)

The redeploy restarts agentbbs, and the smoke test could run before it had
re-bound :22 (0.3 s after on the #136 deploy). One probe made most deploys go
red while the BBS answered moments later. Probe up to 12 times, 10 s apart,
and fail only if :22 never answers. Still a keyscan only, so no account is
ever registered.

Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

.github/workflows/deploy.yml

@@ -162,9 +162,18 @@ jobs:
           # join@ connection would register the connecting key as a new account,
           # so we never complete a handshake here. Admin OpenSSH lives on the
           # admin port, so anything serving :22 is agentbbs answering join@/bbs@.
+          #
+          # Retry: the redeploy above restarts agentbbs, and this step can start
+          # before it has re-bound :22. A single probe failed most deploys
+          # (red runs with the BBS answering fine moments later), so give it
+          # up to ~2 minutes and fail only if it never comes back.
+          for attempt in $(seq 1 12); do
             if timeout 15 ssh-keyscan -T 10 -p 22 "$DEPLOY_HOST" 2>/dev/null | grep -q .; then
-            echo "::notice::agentbbs is serving SSH on ${DEPLOY_HOST}:22 (join@ is reachable)"
+              echo "::notice::agentbbs is serving SSH on ${DEPLOY_HOST}:22 (join@ is reachable, attempt ${attempt})"
-          else
+              exit 0
-            echo "::error::nothing is serving SSH on ${DEPLOY_HOST}:22 — agentbbs may be down"
-            exit 1
             fi
+            echo "attempt ${attempt}/12: nothing on ${DEPLOY_HOST}:22 yet, retrying in 10s"
+            sleep 10
+          done
+          echo "::error::nothing is serving SSH on ${DEPLOY_HOST}:22 after 12 attempts — agentbbs may be down"
+          exit 1