[llvm] [Offload] Add olGetKernelMaxGroupSize (PR #142950)

[Ross Brunton via llvm-commits]

RossBrunton wrote:

I want to expose the number directly to UR (I guess so users can try out different values for local memory and see which works best). I don't think having a launch option for "determine the best option" would fix this (although I don't disagree with the existence of such a function entirely).

It's currently specific to Nvidia, but an implementation can be made for AMD (as it is in HIP) and users will probably appreciate having it available in the liboffload library rather than having to roll it themselves.

https://github.com/llvm/llvm-project/pull/142950