Replace old tensor slicing methods with new API

[robertknight]

Remove the old TensorBase::{slice, slice_mut, slice_dyn, slice_mut_dyn} implementations and replace them with the slice_with APIs that infer the result layout type based on the input layout and slice range.

This now leaves the API with a smaller and easier to use set of methods for slicing, which automatically infer the output view's layout:

// Slice static or dynamic rank tensor, return a view that is
// static or dynamic rank respectively.
tensor.slice((.., 1, 2))

// Same, but returns a result
tensor.try_slice((.., 1, 2))

// Same, but returns a mutable view
tensor.slice_mut((.., 1, 2))

// Same, but returns a mutable view in a result
tensor.try_slice_mut((.., 1, 2))