Honestly I believe modeling such thing in JS as an initial iteration would have drastically changed what ReactJS is today, and likely for the worse. No proof of that, aside from the fact that iterating it in SML gave the rather paradigm-shifting ReactJS we know today.

Can't speak for Jordan but I'm guessing that the answer is "it just felt right". So you can consider ReactJS as a manual program extraction from SML.

I'm not sure whether the original one was ever fully finished. I think there was a blocker from the React reconciler that SML couldn't solve in its type system? It needed GADT and existential, which is why he moved onto OCaml. Related: https://drup.github.io/2016/08/02/difflists/

So in a sense, we knew for a while now that ML would work out, somewhat demonstrated by the real-world usage of React. The technical merits are there; the social ones, definitely not. Thus Reason, which you can view as ReactJS finally going back to its root through ReasonReact, and with the social aspect properly being worked on.