Scheme was developed from LISP to prove the possibility of constructing a language from the ground up using a Meta-circular evaluator. That was important for SICP because it meant that every student knew how to create a language using arbitrary primitives, a DSL. There is no way in an 1 year introductory Python class the students would be anywhere near ready to implement a Python with different primitives. You can argue that the entire LISP philosophy is the wrong approach to solving complex problems, and I think history has perhaps show that, but arguing it was just stupid is a bit much.
As for going 18 levels deep that probably should be broken out about every 5 levels into smaller simpler functions.
As for arrays:
(array dimensions element0 ...) -> array
(array '(2 3) 'a 'b 'c 'd 'e 'f) -> {Array 2 3}
That doesn't seem that hard.
as a functional language, new students must first learn to think along a different paradigm, one opposed to how they had been thinking their whole life, before they can begin to comprehend the basic concepts they're supposed to be learning
Most students can handle Excel, a functional language. I think breaking them of their bad habits is a good thing for an intro course. It puts most of the students on the same level regardless of background. In a week they won't get there, in a year certainly.
Mostly if I were teaching an intro course today I'd probably go with Python. But I don't think it is nearly as clear as you do. Haskell for example (which keeps most of the essence of LISP with giving them more modern concepts) would be a consideration.