BLD: switch to Cython 3.0, forbid deprecated Numpy C API in generated code

[neutrinoceros]

No description provided.

[matthewturk]

my guess is that we'll see pretty enormous performance regressions here if we aren't careful

[neutrinoceros]

Yes, that's what I'm worried about too. Is there a way to test for it that doesn't involve running yt's test suite ?
Maybe we could just systematically change nogil into noexcept nogil ?

[matthewturk]

I think that's a good idea. It might be sufficient just to do some SPH queries with this to demonstrate the slowndown; I bet they will show up in larger runs, but not smaller ones.

[neutrinoceros]

Actually nogil only shows up on one line

ewah_bool_utils/ewah_bool_utils/ewah_bool_wrap.pyx

Lines 23 to 24 in 36a0072

	cdef extern from "<algorithm>" namespace "std" nogil:
	Iter unique[Iter](Iter first, Iter last)

I confess I do not understand this definition. Is it critical ?

[neutrinoceros]

ping @matthewturk

[neutrinoceros]

I'm thinking we could do a release candidate to enabling testing on yt without compromising performance on production for anyone.