DEP: what to do about Numpy 2.0.0 final

[neutrinoceros]

Here's what we know about recent and upcoming events in the ecosystem

• NumPy 2.0.0 is at the release candidate stage, so the final, user-visible release should be a couple weeks to months away now
• because it's an ABI break, every package that depends on it and ships compiled code needs a new release
• we never pinned numpy <2.0
  As a result, when 2.0.0 final comes out, installing nonos in a new env will by default install numpy 2.0 unless something else in our dependency tree pins it. This is not a problem if everything in our dependency tree is already compatible at that point, but it might not be the case:
  I see in particular that the first version of scikit-image (which we depend on via lick) with numpy 2 compat will not support Python 3.9, while numpy 2.0 supports it. I'd conclude that pip install nonos in a new env might create a broken env on Python 3.9, which will be visible if anyone tries to use NonosLick in these conditions.

The easiest solution on our side would be to drop support for 3.8 and 3.9 but it's a very aggressive move. We could also mitigate the issue by pinning numpy<2 ; python_version < '3.10', but should it be done in nonos or in lick, who's the one bringing the problematic dependency ?

Here's a tracker issue to see the whole ecosystem moving towards numpy 2
numpy/numpy#26191

[neutrinoceros]

fixed by removing scikit-image as a dependency in lick 0.6.0