Since I've got a stray downvote, let me clarify: I'm pro-FP. Purity amounts to referential transparency which is uncomplicated. Haskell and similar languages have the character of arithmetic expressions (no side effects) evaluated mechanically. This claim is uncontroversial and is supported by the well known papers of Strachey and Landin. Monads are useful but their use does represent a decision to bring "the awkward squad" of side effects etc. into the clean elegant world of calculating the value of expressions. Alternative approaches might see imperative code written separately from pure code.
It's surely a category error to say that purity is more essential than impurity. What's pure is the expression one types in, and its various reduced forms, culminating in a value. Impure is the messy causal world of real hardware and objects which have stubborn or mutable properties: the whole address-based system of letters in postboxes, for example. The world has state. That doesn't make the platonic ideal of an arithmetic expression any less real, but it's an abstraction of a certain circumscribed set of operations that we perform, typically on a blackboard or with pencil and paper. It's a tiny part of reality.
It's surely a category error to say that purity is more essential than impurity. What's pure is the expression one types in, and its various reduced forms, culminating in a value. Impure is the messy causal world of real hardware and objects which have stubborn or mutable properties: the whole address-based system of letters in postboxes, for example. The world has state. That doesn't make the platonic ideal of an arithmetic expression any less real, but it's an abstraction of a certain circumscribed set of operations that we perform, typically on a blackboard or with pencil and paper. It's a tiny part of reality.