Weak baselines and reporting biases lead to overoptimism in ML for fluid PDEs