DEP: Deprecate error clearing for special method in array-coercion

[seberg]

When looking up __array__ and __array_interface__, previously
most errors were ignored (during the attribute lookup for array).

This deprecates this, in the future only AttributeError (which is
ignored in LookupSpecial itself) will be ignored.

Closes gh-18723

---

Closes gh-18199 I feel like there is probably at least one more issue open, but didn't find it.

[eric-wieser]

Suggested change

	To silence the warning, the type raising a warning has to be adapted
	To silence the warning, the type raising the exception has to be adapted

[eric-wieser]

Looks sensible, although it might be better to use the word "exception" rather than "error in a few places.

[seberg]

Well, that was confusing :). I changed the error message (error -> exception), so the warning filter was not set up correctly in the test and thus the strange failure.

Should probably just squash merge this...

[charris]

Thanks Sebastian.

[radarhere]

Thanks. Just out of interest, is there a standard numpy timeline on moving from deprecation to exception?

[seberg]

@radarhere typically 2 releases (EDIT: 2 releases means 1 year), although its not uncommen we forget, but after two releases we will definitely merge a PR unless there is some protest. If there are very good reasons, I guess we could speed things up after some discussion. (In this case, the only known issue was because shapely ran into this change, since shapely is fixed, I might consider it. But it should still have a good reason of why it helps.)