Pragmatic, sloppy languages will always be more popular, because they are more forgiving.
I think the issue you have stems from the fact that people can't transfer the knowledge from C/C++/Python/wtv directly. One of the standard way to build programs in Haskell (mtl-style) allows you to perform IO actions almost anywhere, provided you're willing to play with monads. But this is so different from anything most people have ever seen that I understand why it can be frustrating.
I have picked up three well-regarded Haskell books over the past decade trying to get back into the language, and don't recall any mention of this capability.
That's another issue, the dearth of anything like tutorials or books on how to do something practical with Haskell, rather than the whirlwind tour of language features. I keep being told that serious software is being built in Haskell, but the knowledge of how to get from university-grade code to production-grade code is an unmarked wilderness that everyone must apparently navigate on their own.
I'm a big fan of the "Haskell Programming from First Principles" book. I read it on-and-off over the course of 2 years, and I think I'm an intermediate Haskell user at this point.
The most production-grade Haskell software I've ever played with would be Pandoc (document conversion) and the Yesod family of webdev libraries. Playing with Yesod is a bit scarier because it uses some Template Haskell magic. On the flip side, there's a great user guide (the "Yesod book").
But you absolutely _can_ do this in Haskell.
Sure, it's considered very unsafe and shouldn't be used in production, but for printf debugging it's fine.
Admittedly, production-ready logging requires type signature modification, but if you subscribe to Haskell's idea that side-effects should be reflected in type signatures, then I don't see that as excessive.
You're mistaken about printf debug.
I tend to use Scala and Java for my day job. I completely understand you. Though I feel that a lot of what's good in Scala is better and simpler in Haskell, or was influenced by Haskell, so even if I don't use it for my day job, I feel it has made my job easier and better. Knowing Haskell has made me a better Scala programmer.
As an example, whenever I had trouble visualizing what some functions would do in Scala -- because they used to have horrible signatures, though it's getting better -- I would simply test the equivalent functions in GHCi to get a feel of what they would do, because Haskell functions tend to be simpler than Scala's ;)
The "over-zealous axioms" you refer to can be boiled down to: if you do mutation in a function, don't give it a pure type.
https://hackage.haskell.org/package/monad-ste Is a fun imperative data structures with abort only exception lib I wrote a while ago that I’m still quite proud of