Rework subview methods and other related methods

[jturner314]

This PR does the following:

• Rename subview_inplace to collapse_axis (deprecating the old name).
• Rename subview_mut toindex_axis_mut (deprecating the old name).
• Rename into_subview to index_axis_move (deprecating the old name).
• Rename private function do_sub to do_collapse_axis.
• Rename subview to index_axis (deprecating the old name).
• Improve docs.
• Rename slice_inplace to slice_collapse (deprecating the old name).
• Add insert_axis_inplace and index_axis_inplace for IxDyn arrays.
• Deprecate remove_axis.

These changes make method names more uniform, especially between subviews and slicing, and provide new functionality for IxDyn arrays.

Possible alternative names for "collapse" include "narrow" and "restrict".

After a while with slice_inplace deprecated, we can remove it, then add a slice_inplace method for IxDyn arrays that removes axes like slice_move does.

What do you think?

[bluss]

Nice reorganization.

[bluss]

Great. I don't have the release overview at the moment, if we have breaking changes in flight already or not. Is it time to start merging them? I can make point releases before then if we need to.

[jturner314]

I don't have the release overview at the moment, if we have breaking changes in flight already or not.

I just submitted #540, which describes the changes between 0.12.0 and the current master. We haven't merged any breaking changes yet.

[jturner314]

One point of interest about the way I've implemented insert_axis_inplace and index_axis_inplace in this PR: the dim and strides are temporarily inconsistent. This is fine, though, even if self.strides.insert_axis(axis) or self.strides.remove_axis(axis) panics, because nothing can observe the inconsistency since ArrayBase doesn't implement Drop and &mut ArrayBase<S, D> is !UnwindSafe.

On a related note, I'm currently in the process of clarifying the invariants that methods on ArrayBase must uphold (and when the invariants need to be upheld) and updating the array creation methods to enforce the invariants. (This will also fix a couple of edge cases that currently result in undefined behavior.) I'll submit a PR when I'm done.

[jturner314]

I just rebased the commits off the latest master and updated the array0_into_scalar test added in #535.

[jturner314]

Is this ready to merge? Everything looks good to me.

[bluss]

Sure!