[llvm] [AMDGPU] always emit a soft wait even if it is trivially ~0 (PR #147257)

[Sameer Sahasrabuddhe via llvm-commits]

ssahasra wrote:

Refined the change to narrow down exactly how it is that we want to use the S_WAITCNT_soft ... the memory legalizer now inserts a ~0 wait only for operations that involve a release, at a scope that is workgroup or higher. This drastically reduces the churn visible in the tests while making the intention more obvious.

https://github.com/llvm/llvm-project/pull/147257