Currently, views-of-views construct their re-indexed indices by slicing
into the parent indices. This PR changes this to use views of the parent
indices instead. This makes the operation faster and non-allocating if
the `parentindices` for the original view are `Vector`s.

```julia
julia> a = rand(200, 200);

julia> av = view(a, collect.(axes(a))...);

julia> @btime view($av, axes($av)...);
  312.393 ns (4 allocations: 3.25 KiB) # master
  7.354 ns (0 allocations: 0 bytes) # PR
```
Indexing into the resulting view seems equally fast in simple cases:
```julia
julia> av2 = view(av, axes(av)...);

julia> @btime sum($av2);
  66.883 μs (0 allocations: 0 bytes) # master
  66.888 μs (0 allocations: 0 bytes) # PR

julia> av2 = view(av, collect.(axes(av))...);

julia> @btime sum($av2);
  66.886 μs (0 allocations: 0 bytes) # master
  66.891 μs (0 allocations: 0 bytes) # PR
```

---------

Co-authored-by: N5N3 <2642243996@qq.com>