[CIR] Fix HLSL test that crashes (#182894)
This was caused by #182609, which just changed the way the AST stores
these, which causes us to hit an NYI in a way that doesn't recover
nicely. In the future, we could probably represent a 'no op' instead of
an empty op in the IR for these cases, but there isn't much use for it,
since it is always after NYI.
This patch changes the test to use float instead of float1 as suggested
in review, which avoids the problematic conversion.
