Add view_as_real support (#1512)
I added two new types `Float_2` and `Double_2` to `DataType`. In the future, we might need to add more vector types if we decide to support full `view(dtype)`.
I also renamed `ViewDtypeOp` into `ViewAsScalar` and changed its definition: now `ViewAsScalar` has one input TV which must have a vector dtype, and this operator will append a new `IterDomain` of type `IterType::VectorComponent` which iter through the vector dimension. For example, if the input TV of `ViewAsScalar` is `TV[iS{i0}]` of type `Float_2`, then the output will be `TV[iS{i0}, vS{2}]` of type float.
`EraseType` is also renamed to `BitCast`.
I also added lowering for `ViewAsScalar` in `IndexLowering`.
