It seems to me that these two statements are at odds (Carvalho even puts it in bold). If there is a first mover advantage where enough miners signal above the 55% threshold before the August deadline can cause the rest of the miners to get caught un prepared and potentially miss out on revenue, then it seems like there is an incentive to signal early...

You're right, but also this dynamic is observed in finitely repeated games regularly. Even though you can infer through backwards induction that the equilibrium strategy is to defect (signal) immediately, real human players often play out a bunch of rounds before defecting.