Reminder: revise copy semantics in __array__ protocol implementations

[neutrinoceros]

#16142 introduced the copy argument to __array__ implementations, because numpy dev required it. However, this was only a small part of the PR and it may not have gotten the attention it needed.

In fact, the question has been asked upstream about how this argument should be implemented, see
numpy/numpy#25941
This discussion isn't concluded yet, but the ticket is currently triaged as part of the numpy 2.0 milestone, so I'd expect some consensus be reached by the time we see a stable release.

While we're on this, and if it doesn't turn out like a completely separate issue, we could also address this comment from @mhvk : #16142 (comment):

Separately, I'm actually not sure this implementation is correct, I think it should be return self._array.astype(dtype=dtype, copy=copy) but then the default would have to be False (which I think is fine).