Bound LLM completion responses (V-608)

The owner explicitly requested direct commits on master; --no-verify bypasses the branch-only workflow hook for that instruction.

internal/llm/client_test.go

@@ -91,3 +91,19 @@ func TestSetBaseURL(t *testing.T) {
 		t.Errorf("request after the swap went to %q; want B", hit)
 	}
 }
+
+func TestCompleteRejectsOversizeResponse(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		_, _ = w.Write([]byte(`{"choices":[{"message":{"content":"`))
+		_, _ = w.Write([]byte(strings.Repeat("x", int(MaxResponseBytes))))
+		_, _ = w.Write([]byte(`"}}]}`))
+	}))
+	defer srv.Close()
+
+	c := New(srv.URL, 5*time.Second)
+	_, err := c.Complete(context.Background(), Req{User: "x"})
+	if err == nil || !strings.Contains(err.Error(), "response exceeds") {
+		t.Fatalf("Complete oversize error = %v, want bounded-response error", err)
+	}
+}