BUG: Fix incorrect refcounting in new asarray path

[seberg]

The new path to preserve dtypes provided by creating a view got the reference counting wrong, because it also hit the incref path that was needed for returning the identity.

This fixes up gh-21995

Closes gh-22233

@charris I keep forgetting how (should note it somewhere :)). Could you retrigger the nightlies, since this seems to affect downstream CI quite annoyingly.

[seberg]

(I did not add a test, mem-leak testing would have found it easily, we just don't run it regularly. And it is so massive, that everyone notices anyway :))

[rgommers]

@charris I keep forgetting how (should note it somewhere :)). Could you retrigger the nightlies, since this seems to affect downstream CI quite annoyingly.

I removed the broken wheels uploaded in the last ~30 hours. For the record, this is how:

$ anaconda login  # need credentials that access the bucket
$ anaconda show scipy-wheels-nightly/numpy  # shows all releases
$ # pick the right version (double check on web UI)
$ anaconda remove scipy-wheels-nightly/numpy/1.24.0.dev0+750.ge9295bf99

[rgommers]

Thanks for the quick fix @seberg!

[mattip]

One can also remove the wheels using the gui at https://anaconda.org/scipy-wheels-nightly/numpy/files if logged in and permitted.

To run the wheel CI

• go to Actions at the top of the page -> Wheel builder (on the left under "workflows") -> "Run workflow" button in the blue area at the top of the table. This will run the x86 and macOS wheel builds
• go to https://app.travis-ci.com/github/numpy/numpy -> options (hamburger menu on the right) -> trigger build which will build wheels for aarch64 and others.

[seberg]

Ah, thanks! I likely don't have rights to yank the wheels from anaconda, but there is no need for that anyway. However, triggering a new build should work for me probably.