Support Tensor-Scalar operations

[hw-yuexy]

In MLIR, linalg.add (for example) can take a tensor and a scalar, and add the scalar to every element of the tensor. We should relax the condition in verify_memref_tensor_types to support this kind of operation.

[Balint-R]

Are you sure it's linalg.add? I thought only arith.add allows mixing tensor and scalar.

[hw-yuexy]

Yeah from the documentation, the input can be "variadic of any type", so we can actually pass in a tensor and a scalar.

[Ritsuka314]

Doc says: The shapes and element types must be identical. The appropriate casts, broadcasts and reductions should be done previously to calling this op.