Fix asarray(copy=False)

[asmeurer]

For NumPy 2.0, this is implemented directly. For NumPy 1, we emulate it by checking if asarray() creates a copy or not.

This also removes support for the np._CopyMode enum in asarray(), as this is not portable.

@rgommers I can't remember if there was a good reason that we didn't just emulate copy=False in this way originally (instead, we raised NotImplementedError).

[rgommers]

Thanks @asmeurer, this looks good to me.

I don't think there was any reason to not implement this beyond no one thinking about the option to try with copy=False and then checking if a copy was made afterwards, like you are doing in this PR.

[asmeurer]

Maybe it had to do with when obj is not an array. Are there instances where copy=False could fail with the buffer protocol?

[rgommers]

Maybe it had to do with when obj is not an array. Are there instances where copy=False could fail with the buffer protocol?

No, neither the buffer protocol nor __array_interface__ have that.

[asmeurer]

Searched through the discussion at numpy/numpy#18585 and couldn't find anything. I think I just wasn't really clear at the time that numpy copy=False worked like copy=None.

[asmeurer]

OK, I updated this to handle Python built-ins and buffer protocol objects properly, and added some more tests since I don't think the test suite really checks copy that sufficiently.

[rgommers]

Looks good, let's get this in. Thanks Aaron.