BUG: common: fix cholesky upper decomp for complex dtypes

[lucascolley]

Closes gh-54.

cc @asmeurer

[rgommers]

Thanks @lucascolley! I verified the change on your NumPy PR, and this matches that one. In it goes.

[rgommers]

Actually, let me re-run CI since a PyTorch job was red and I couldn't see the CI log anymore.

[lucascolley]

thanks Ralf. I wonder whether any changes are needed to array-api-tests since this wasn't caught initially?

[rgommers]

I wonder whether any changes are needed to array-api-tests since this wasn't caught initially?

I checked quickly, and yes indeed - upper is tested, but only with real-valued dtypes (via floating_dtypes from hypothesis). I guess the fix is to use complex_dtypes from https://hypothesis.readthedocs.io/en/latest/_modules/hypothesis/extra/array_api.html. Are you interested to follow up on that?

[rgommers]

Several failures due to fft testing mainly, I assume because new tests were added in the test suite and not yet here.

All the linalg.cholesky tests passed, so I'll go ahead and merge this.

[lucascolley]

Are you interested to follow up on that?

I had a look but I'm not sure how to fix. Will open an issue instead.