Use find_package(Python) instead of PythonInterp

[VincentRouvreau]

Modification taken from #151
Fix #139
CMake 3.12 is now 5 years old 3.15 is now 4 years old (july 2019), so it should be ok for every one.
CMake 3.15 is required for cmake_policy(SET CMP0094 NEW)
CMake 3.16 (nov 2019) is required if the user wants to specify a python interpreter (cmake -DPython_EXECUTABLE=/custom/path/to/python3 ...)

[mglisse]

Why do we need that?

[VincentRouvreau]

I precised it on 7e33f42 because macosx-11.7-universal2 has several python installation available and cmake was not finding the correct one, cf. https://github.com/GUDHI/gudhi-devel/actions/runs/7030598162/job/19130456933#step:5:60
When I precise Python_EXECUTABLE, the error is no more.
I wanted to do the same for Windows, but my batch command seems not correct, WIP

[mglisse]

I guess my question is whether users are likely to be hindered by this change in our CMake scripts (just like we were affected in our CI), and how we could help. For instance, if cmake happily stops when it finds a version of python that doesn't have the development parts, we could tell find_package that we require components Interpreter and Development.Module (and possibly NumPy) so hopefully it keeps looking for another python in that case. There are also several options that can affect the strategy:

• setting policy CMP0094 to NEW changes Python_FIND_STRATEGY a bit
• Python_FIND_UNVERSIONED_NAMES can be set to FIRST if we prefer python to python3.X

There are also Python_FIND_REGISTRY (windows) and Python_FIND_FRAMEWORK (macos), but they don't look like things we should force on users. We could use them in CI, but in CI specifying the executable (as you did) is probably more reliable.

By the way, does Python_EXECUTABLE need a full path, or does -DPython_EXECUTABLE=python work?

[mglisse]

By the way, does Python_EXECUTABLE need a full path, or does -DPython_EXECUTABLE=python work?

https://cmake.org/cmake/help/latest/module/FindPython.html says

Note: All paths must be absolute. Any artifact specified with a relative path will be ignored.

[VincentRouvreau]

I didn't notice it the first time I wrote the PR, but setting policy CMP0094 to NEW seems promising as it is using the python interpreter from my conda environment instead of the one from my system. It worth it to test it, even if it requires cmake 3.15 (july 2019). I rewrite some python cmake horrors and I will give it a try