feat: Implement issorted for AbstractGPUArray without scalar indexing.

[shreyas-omkar]

Fix for #635

Base.issorted on AbstractGPUArray currently falls back to scalar iteration, which triggers scalar indexing errors and prevents correct usage on GPU-backed arrays.

This PR introduces a GPU-native implementation of issorted using a kernel-based approach, ensuring the operation executes entirely on the device.

Tests

(GPUArrays with base as test_args)

---

(GPUArrays)

[shreyas-omkar]

@kshyatt please review.

[maleadt]

IIUC Ordering types are supposed to implement lt(ord, a, b), so we could probably use that?

[shreyas-omkar]

I think Base.Order.lt is not GPU-safe. It calls methods that throws. I'm getting jl_f_throw_methoderror if used. Am I missing something??

[kshyatt]

Thanks for getting to this so quickly! I'm at a conference today so may not get to it immediately but perhaps @maleadt has spare time? If not I'll look on Monday.

[shreyas-omkar]

@maleadt I have removed the unnecessary using KernelAbstractions import as suggested

I initially tried using Base.Order.lt(ord, a, b), but this results in invalid GPU IR.
To avoid this, the current implementation resolves order and rev on the host and passes plain lt / by functions into the kernel, which keeps the device code simple and GPU-safe.

Please let me know if I'm missing something.