On the lognormality of historical magnetic storm intensity statistics: Implications for extreme‐event probabilities