[Codegen] Add lowering functions for extract/slice/insert

[AlexRTer]

In order to run scripts using the --mlir-codegen cli flag with little to no dependence on pre-compiled kernels it would be beneficial to be able to lower extract, slice, and insert operations to other MLIR dialects. In many cases, this can help to avoid converting a DenseMatrix to a MemRef and back between operations that support codegen and those that do not. The three mentioned operations are very common and should be straight-forward to implement.

A possible approach is to lower these to a Linalg GenericOp with (one of) the input MemRef being the result of a MemRef SubViewOp. An example of this combination can be found in src/compiler/lowering/AggDimOpLowering.cpp. SubViewOp takes Offsets, Sizes, and Strides as arguments which can easily implement a view on a single row for example, by setting offset: (1,0), sizes: (1, numCols), strides: (1,1) to create a view on the 2nd row of a MemRef if I am not mistaken.