feat(update): auto-rollback OTA binaries that crash-loop before they can self-verify

.198 crash-looped 236x on a truncated OTA binary with a good backup
sitting in update-backup/ — verify_pending_update() can't help when
the new binary never runs. New scripts/ota-crash-guard.sh runs as
root from ExecStartPre: while the post-OTA pending-verify marker
exists it counts start attempts, and after 5 failures restores the
backup binary (atomic rename), replaces the marker with an
update-rolled-back.json tombstone, and lets the service come back
on the previous version.

Wiring: fresh ISOs get the ExecStartPre line in the unit file;
existing nodes get a drop-in installed by apply_update's runtime
component step on their next OTA. '+-' prefix so a missing or
failing guard can never block the service.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

scripts/ota-crash-guard.sh

@@ -0,0 +1,72 @@
+#!/bin/bash
+# OTA crash-loop guard — runs as root from ExecStartPre=+- on archipelago.service.
+#
+# Covers the failure mode verify_pending_update() cannot: a freshly-applied
+# binary that can't even start (SEGV/ENOEXEC — e.g. the truncated 17MB binary
+# .198 installed on the v1.7.103 OTA, which crash-looped 236 times with a
+# perfectly good backup sitting in update-backup/). The in-binary probe never
+# runs because the binary never runs, so this guard counts start attempts from
+# outside and restores the backup binary once the new one has clearly failed.
+#
+# Scope is deliberately narrow: it acts ONLY while the post-OTA pending-verify
+# marker exists (written by apply_update just before the restart, deleted by
+# the new binary once it boots and passes its probes). A crash loop with no
+# marker is not an OTA gone wrong, and this script stays out of it.
+#
+# Always exits 0 — a guard must never be the reason the service can't start.
+
+set -u
+
+DATA_DIR=/var/lib/archipelago
+MARKER="$DATA_DIR/update-pending-verify.json"
+COUNT_FILE="$DATA_DIR/ota-crash-guard.count"
+BACKUP="$DATA_DIR/update-backup/archipelago"
+BINARY=/usr/local/bin/archipelago
+MAX_ATTEMPTS=5
+
+log() {
+    echo "$*" | systemd-cat -t ota-crash-guard -p warning 2>/dev/null || true
+}
+
+# No pending OTA verification -> nothing to guard; clear any stale counter.
+if [ ! -f "$MARKER" ]; then
+    rm -f "$COUNT_FILE"
+    exit 0
+fi
+
+# Count this start attempt. The counter only accumulates while the marker
+# exists; a healthy new binary deletes the marker on its first successful
+# boot, and the next start clears the counter above.
+count=$(cat "$COUNT_FILE" 2>/dev/null || echo 0)
+case "$count" in ''|*[!0-9]*) count=0 ;; esac
+count=$((count + 1))
+echo "$count" > "$COUNT_FILE" 2>/dev/null || true
+
+if [ "$count" -lt "$MAX_ATTEMPTS" ]; then
+    exit 0
+fi
+
+if [ ! -f "$BACKUP" ]; then
+    log "OTA crash guard: $count failed start attempts but no backup binary at $BACKUP — cannot roll back"
+    exit 0
+fi
+
+# Already restored (or the OTA never replaced the binary)? Don't loop.
+if cmp -s "$BACKUP" "$BINARY"; then
+    exit 0
+fi
+
+# Restore via copy-to-temp + atomic rename; never truncate the live path.
+tmp="$BINARY.rollback.$$"
+if cp "$BACKUP" "$tmp" && chown root:root "$tmp" && chmod 755 "$tmp" && mv "$tmp" "$BINARY"; then
+    # Leave a tombstone for the UI/logs instead of the marker so the restored
+    # binary doesn't run the post-OTA probe against the rolled-back version.
+    mv "$MARKER" "$DATA_DIR/update-rolled-back.json" 2>/dev/null || rm -f "$MARKER"
+    rm -f "$COUNT_FILE"
+    log "OTA crash guard: restored previous binary after $count failed start attempts of the updated one"
+else
+    rm -f "$tmp" 2>/dev/null
+    log "OTA crash guard: failed to restore backup binary (cp/mv error)"
+fi
+
+exit 0