fix(error-classifier): quota/billing errors are non-retryable STOP (fixes #3126)

Add STOP_MESSAGE_PATTERNS that take precedence over RETRYABLE_MESSAGE_PATTERNS.
Message-only quota errors (no error name) now correctly return false from
isRetryableModelError, preventing unnecessary fallback retries on exhausted quotas.

- quota will reset after...
- quota exceeded
- usage limit has been reached
- free usage limit / billing limit / plan limit / subscription limit
- out of credits / credits exhausted / insufficient credits / insufficient balance

Also add 4 regression tests covering message-only quota STOP cases.

src/shared/model-error-classifier.ts

@@ -40,8 +40,6 @@ const RETRYABLE_MESSAGE_PATTERNS = [
   "rate_limit",
   "rate limit",
   "quota",
-  "quota will reset after",
-  "usage limit has been reached",
   "all credentials for model",
   "cooling down",
   "exhausted your capacity",
@@ -76,6 +74,25 @@ const RETRYABLE_MESSAGE_PATTERNS = [
   "529",
 ]
 
+/**
+ * Message patterns that indicate a non-retryable STOP error (quota/billing exhaustion).
+ * These take precedence over RETRYABLE_MESSAGE_PATTERNS.
+ */
+const STOP_MESSAGE_PATTERNS = [
+  "quota will reset after",
+  "quota exceeded",
+  "usage limit has been reached",
+  "free usage limit",
+  "billing limit",
+  "monthly limit",
+  "plan limit",
+  "subscription limit",
+  "out of credits",
+  "credits exhausted",
+  "insufficient credits",
+  "insufficient balance",
+]
+
 const AUTO_RETRY_GATE_PATTERNS = [
   "rate limit",
   "quota",
@@ -121,6 +138,12 @@ export function isRetryableModelError(error: ErrorInfo): boolean {
 
   // Check message patterns for unknown errors
   const msg = error.message?.toLowerCase() ?? ""
+
+  // STOP patterns take precedence over retryable patterns
+  if (STOP_MESSAGE_PATTERNS.some((pattern) => msg.includes(pattern))) {
+    return false
+  }
+
   if (hasProviderAutoRetrySignal(msg)) {
     return true
   }