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

[Matt Arsenault via cfe-commits]

arsenm wrote:

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

It should be calling in on every single function 

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