KA.jl-related slowdowns

The switch to KA.jl significantly slowed down several operations.

---

CUDA.jl: `permudetims`, `broadcast`, and many others

https://speed.juliagpu.org/changes/?tre=10&rev=6221589f5befec8f6f157a5a5271667dba09d0b6&exe=11&env=1

---

Metal.jl: `permudetims`

```
private array/permutedims/4d 2911500 ns 860084 ns 3.39
private array/permutedims/2d 1065021 ns 862229.5 ns 1.24
private array/permutedims/3d 1629229 ns 919520.5 ns 1.77

shared array/permutedims/4d 2933000 ns 858875 ns 3.41
shared array/permutedims/2d 1054250 ns 862292 ns 1.22
shared array/permutedims/3d 1625958 ns 923916.5 ns 1.76
```