[NFC] Simplify WaveActiveMax test

[llvm-beanz]

This was a test I had written to cover some edge cases, and the test is a bit awkward and buggy. This corrects the test to hopefully be more consistent across vendors.

[farzonl]

is EdgeConditions a temporary stop gap? Should we be working with Apple to drive conformance?

I know right now this is just for Metal, and for this particular case the DirectX driver matches spirv/vulkan, but are their Vulkan cases we should be considering for EdgeConditions aswell for other HLSL intrinsics? If so are we tracking them somewhere to burn down the list? This pr doesn't have a ticket associated with it so want to know what the long term plan is of adding a new cmake build construct like OFFLOADTEST_ENABLE_EDGE_CONDITIONS.

[bogner]

I'm not convinced adding a whole extra configuration here is worth it, so I would suggest dropping the EDGE_CONDITIONS cmake variable. More ways to test just means we'll have a harder time convincing ourselves we have proper coverage.

The test simplifications look good to me though - I suspect avoiding overwriting the same values in multiple threads will make this test a lot less flaky.

[bogner]

This logic is exactly backwards