Review treatment of `__array__` attribute

I'm writing this issue from memory after finding an old note to myself that I wrote in response to a conversation I had with @tlambert03 some time last year. 😅

* The `imagej.images.is_arraylike` function checks for four attributes: `shape`, `dtype`, `__array__`, and `ndim`. Is this correct?

* The classes that augment Java-side `net.imglib2.RandomAccessibleInterval` and related objects to be more arraylike include implementations of `shape`, `dtype`, and `ndim` on the appropriate superinterfaces, but nothing implements `__array__`. Should it?

Ideas and opinions welcome!

Start with imagej.images.is_arraylike and inspect the classes augmenting Java-side net.imglib2.RandomAccessibleInterval and related objects. Compare their shape, dtype, and ndim implementations with the function's __array__ check, then seek a maintainer decision on the intended treatment. Done means the behavior and any required change are agreed and covered appropriately.

Written by the indexing model from the issue text.