[clang] clang/AMDGPU: Set amdgpu-max-num-workgroups to disable Y/Z by default (PR #119009)

[Shilei Tian via cfe-commits]

shiltian wrote:

> The kernels do, but the callable __omp_offloading functions are missing them for some reason.

The OpenMP GPU CodeGen only calls it on the target entry point.

https://github.com/llvm/llvm-project/pull/119009