Add basic cuda/hip API interface

API's exposed to end user (`kaolin_array`)
* Magic `=` for fortran allocatable to/from kaolin array
* `copy_from` (return fortran allocatable)
* `copy_to`

GPU Memory management
* `gpumalloc`
* `gpumallochost`
* `gpufree`
* `gpumemcpy`
* `gpumemcpy`

Data initializers
* `zeros`
* `rand`
* `ones`
* `eye`

For hand-written kernels and kernel wrappers, let's use chevron syntax for kernel launches `<<>>`

Tasks
- [ ] #2
- [ ] ??

Start by reviewing the proposed kaolin_array API, GPU memory-management functions, data initializers, and the reference to task #2. The issue does not name files, entry points, tests, or acceptance criteria, so the scope and definition of done need to be established before implementation.

Written by the indexing model from the issue text.