feat: faster splitAt

[kim-em]

It turns out using List is faster than Array in the auxiliary functions for splitAt and splitAtD. It's about a 10% improvement (and much more for splitAtD when there is lots of padding).

I was a little surprised at first, so double checked my timings from #time #eval ... with a hyperfine "lean --run ..." benchmark.

[digama0]

This is not actually that surprising, but I appreciate you getting hard numbers on this, I'd been hoping someone could verify this one way or another. The reason the list version performs competitively is because List.reverse works in-place, so it's not really wasting any allocations not involved in the final output. The array version is more compact but the array allocation is additional to the output.