"Transformation of data" makes a lot of sense if you're accustomed to "pipes and filters" architectures. It's anything your program might actually do with data that passes through it. Doesn't matter what it is – it's a black box!
Why should handlers and actions be separated? Because they're different things! This is already standard in web architectures: the router is a collection of handlers and the functions that the router calls out to are actions. I like that actions are separated out from the monolithic "controller".