Add support for alignment to cuda array helpers

[tpn]

This issue captures the work discussed during the meeting on Friday 21-Feb-2025, regarding specification of a desired alignment when creating shared memory, e.g.:

temp_storage = cuda.shared.array(dtype=np.int32, shape=N, alignment=64)
thread_data = cuda.local.array(dtype=np.int32, shape=M, alignment=128)

This work will entail Numba changes; I'll link in any issues/PRs from Numba as they get created.

[leofang]

FYI numpy/numpy#28318 (comment) might be relevant once we hear back from Sebastian.

[tpn]

Yeah just reviewed that issue, I'll keep my eye on it. Thanks for tagging me @leofang!

[tpn]

First pass implementation WIP PR is here: numba/numba#9942