Need for specific scipy version installation

[chaimain]

@moralejo found an import error for QhullError from scipy.spatial from the new merger of PR #711 https://github.com/cta-observatory/cta-lstchain/blob/master/lstchain/high_level/interpolate.py#L21 while having scipy 1.7 installed in the conda environment. We do not have any constraints on scipy version currently. This might be essential for the forthcoming major release with major updated dependencies, to avoid any such issues for individual users.

The particular import worked on my system which has scipy 1.9.3 and it also did not create any warnings for this import in the CI.

I tried to use scipy~=1.8 but it had various inconsistencies with a list of other dependencies (pydantic, ctapipe, gammapy, iminuit, etc). With this change, it allows me to make the above import of QhullError but the issue with other dependencies is a problem.

[gabemery]

Is it possible to install scipy 1.7 with current dependencies? If everything worked with 1.9.3 why did you try 1.8?

[chaimain]

Using scipy 1.7 does work with the current dependencies. I had not noticed any issue with the scipy version in my lst-dev conda environment as there was no error so far, and so I am not sure when 1.9.3 was installed exactly.
I wanted to try and see if there is a minimum version that works for the current lstchain, but it seems to not be trivial.

[chaimain]

Working with Gammapy, I just know that there was an issue with scipy v1.10, which was resolved in the bugfix scipy v1.10.1.

[gabemery]

SciPy 1.7.0 : Jun 20, 2021
ENH: make QhullError public (scipy/scipy#15003) : Nov 8, 2021
SciPy 1.8.0:Feb 4th 2022

It seems the change of QhullError implementation which is compatible with your way of importing it is implemented since release 1.8.0. So you should add scipy>=1.8

[chaimain]

It does work and I get scipy 1.9.3 installed and all tests pass. Also checked with the fix in #1092 to be certain. I will make a PR for this change.