[llvm] [AMDGPU] Remove co-exec scheduler diagnostics. (PR #193724)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> Using the CoExecScheduler for processors it doesn't support is not supported, and thus we have the diagnostic. 

I do not want artificial restrictions like this anywhere. 

> By removing this diagnostic we put ourselves in a position where MI450 & MI350 scheduling optimizations are blocked by regressions on gfx900.

This is a case by case decision to evaluate (plus the test coverage for gfx900 is close to nonexistent) 

https://github.com/llvm/llvm-project/pull/193724