[mlir][Shape] Make sure tensor_cast(constant_shape) folding uses the correct type

This is still subtle, but I think the test cases are sufficient to show
that it works.

Differential Revision: https://reviews.llvm.org/D92927