fix(wiggums): exponential backoff on autoLoop halt-watchdog-break

When the halt-watchdog detects stuck state, the autoLoop was logging
"halt-watchdog-break" every iteration but otherwise tight-spinning
through dispatch-resolve at ~2s/iteration. 2026-05-17 dogfood logged
60+ such events in a 30s window — pure CPU burn while the actual
stuck condition stayed stuck.

Fix: exponential backoff (1s → 2s → 4s → 8s → 16s → capped at 30s)
based on how many halt thresholds have elapsed. Heartbeat() resets
when real progress resumes (existing behavior). Backoff costs nothing
when the loop is healthy.

One of the 14 Ralph-Wiggum patterns surfaced this session.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

src/resources/extensions/sf/auto/loop.js

@@ -671,8 +671,18 @@ export async function autoLoop(ctx, pi, s, deps) {
 				iteration,
 				elapsedMs,
 			});
-			// Do not break the loop — the watchdog only emits observability
-			// signals. The operator or a future gate can decide to stop.
+			// #wiggums: when stuck, back off instead of tight-spinning. The
+			// old comment said "Do not break the loop — observability only",
+			// but observability without backoff means we burn CPU writing
+			// halt-watchdog-break events at ~1Hz while whatever's actually
+			// stuck stays stuck. 2026-05-17 dogfood logged 60+ such events
+			// in a single 30s window. Backoff is exponential up to 30s, so
+			// short stalls recover quickly while long stalls don't melt the
+			// CPU. Heartbeat() will clear the backoff when real progress
+			// resumes.
+			const stuckCycles = Math.floor(elapsedMs / DEFAULT_HALT_THRESHOLD_MS);
+			const backoffMs = Math.min(30_000, 1000 * 2 ** Math.min(5, stuckCycles));
+			await new Promise((resolve) => setTimeout(resolve, backoffMs));
 		}
 		// ── Journal: per-iteration flow grouping ──
 		const flowId = randomUUID();