`column` on grids are inconsistent and do not actually make views

https://github.com/CliMA/ClimaCore.jl/pull/1487 fixed some issues, but [`column`](https://github.com/CliMA/ClimaCore.jl/blob/21d1f09f156aa3bdc09bb00889eb816ef26a2f43/src/Grids/column.jl#L42-L43) does not actually make a view, so the underlying DataLayout for a `VIJFH`-backed ExtrudedFiniteDifferenceSpace is still a `VIJFH` DataLayout.

The issue here is that `column` on a broadcasted object now contains a mixture of `VF` and `VIJFH` datalayouts. Memory access patterns for our metric terms are now "stuck" to those governed by `VIJFH`. i.e., we cannot make MArray-backed datalayouts.

This is a hard block, stopping https://github.com/CliMA/ClimaCore.jl/pull/1763 because we cannot make an MArray-backed `VIJFH` (as that would be huge), and we cannot reconstruct `ExtrudedFiniteDifferenceGrid` on the GPU, since it is mutable.

What we have now seems like a bad design in that I would expect `column` to recurse all the way down a broadcasted object, and it doesn't. It's also unfortunate that this is blocking https://github.com/CliMA/ClimaCore.jl/pull/1763.

The same is probably true for `slab`.

Start with src/Grids/column.jl at lines 42–43 and inspect how column propagates through broadcasted objects and DataLayout values. Compare the corresponding slab behavior, focusing on VIJFH-backed ExtrudedFiniteDifferenceSpace and the MArray-backed layout constraint. Done means column, and likely slab, recurse consistently with coverage for broadcasted objects and the affected layouts.

Written by the indexing model from the issue text.