For a new language I would suggest that without annotation, only pure, total functions are allowed.
Haskell makes you annotate side-effecting functions with a tag (like IO). But it allows divergent and partial functions just fine. That's why they can't just remove the laziness.
I propose that the latter would also need a tag.
The nice thing about total functions is that the choice between evaluation orders becomes purely a matter of optimization, and won't change anything about semantics.
Some languages like Agda already do something a bit like this.
(Also note that total functions are not Turing complete. But that's a feature, not a bug.)
What's more Haskell specific is recognizing many problems can be solved easier when seen through the lens of parsers, interpreters and/or compilers.
Haskell's and the whole ML family of languages are great at writing parsers etc. In fact they were originally invented for implementing other languages.
I meant that solving a problem in Haskell can often mean eg writing something a bit like a parser.
Lazy is a ridiculous default, the combinatorial explosion created by GHC extensions makes C compiler quirks look like a gentle massage in comparison, and building Haskell is a friggin migraine on wheels.
But for whatever reason, those folks are on average in my experience, serious as a heart attack.