[llvm] [AMDGPU] Decrease default NSA threshold from 3 to 2 (PR #116624)

[Jay Foad via llvm-commits]

jayfoad wrote:

Also note that we still have GCNNSAReassign which tries to optimize NSA -> sequential form post-regalloc.

https://github.com/llvm/llvm-project/pull/116624