Supported python and numpy versions

[lfarv]

At some point, we will have to revise the python and numpy versions supported in AT. Concerning python maintenance, the present status is the following:

[Python 3.15 (in development)](https://docs.python.org/3.15/)
[Python 3.14 (pre-release)](https://docs.python.org/3.14/)
[Python 3.13 (stable)](https://docs.python.org/3.13/)
[Python 3.12 (security-fixes)](https://docs.python.org/3.12/)
[Python 3.11 (security-fixes)](https://docs.python.org/3.11/)
[Python 3.10 (security-fixes)](https://docs.python.org/3.10/)
[Python 3.9 (security-fixes)](https://docs.python.org/3.9/)
[Python 3.8 (EOL)](https://docs.python.org/3.8/)
[Python 3.7 (EOL)](https://docs.python.org/3.7/)
[Python 3.6 (EOL)](https://docs.python.org/3.6/)
[Python 3.5 (EOL)](https://docs.python.org/3.5/)
[Python 3.4 (EOL)](https://docs.python.org/3.4/)
[Python 3.3 (EOL)](https://docs.python.org/3.3/)
[Python 3.2 (EOL)](https://docs.python.org/3.2/)
[Python 3.1 (EOL)](https://docs.python.org/3.1/)
[Python 3.0 (EOL)](https://docs.python.org/3.0/)
[Python 2.7 (EOL)](https://docs.python.org/2.7/)
[Python 2.6 (EO](https://docs.python.org/2.6/)

Dumping old python versions allows to benefit from new python features, but also, more interestingly, to bump the minimum numpy version to recent versions offering better functionalities (inplace matrix multiplication: @= for instance).

A strategy is proposed as a "community policy standard across the Scientific Python ecosystem" in SPEC 0. According to SPEC 0, we would restrict to python >= 3.11, numpy >= 1.25, scipy >= 1.11.

Keeping support of old versions is not such a burden for us, so it's not necessary to be so strict.

For the time being, I would suggest dropping python 3.7 and 3.8, declared "End of life", and bumping numpy to the minimum supported on python 3.9 (python 3.9 is supported from 1.20 to 2.0). This leaves 5 supported python versions, but python 3.14 is coming soon. This allows removing a few tests in the code referring to python < 3.9.

[oscarxblanco]

Dear @lfarv ,

I agree with your suggestion about dropping python 3.7 and 3.8.
If one needs a version of pyat supporting any of those I think one could go backwards in the pip repository and check the python requirements. Is there any other alternative to it ?

Also, the expired long term release of debian, debian 10, uses python 3.7. And the oldest long term release still supported, Debian 11, uses python 3.9.
https://www.debian.org/releases/index.html
https://wiki.debian.org/Python
https://wiki.debian.org/LTS

[oscarxblanco]

Dear @lfarv ,
I just saw that Python 3.9 is about to end its life cycle (oct/2025) https://devguide.python.org/versions/

Should we deprecate python 3.9 too ?

[lfarv]

@oscarxblanco : good question. I did not see any big change in python between 3.9 and 3.10. This raises the minimum numpy version to 1.22 instead of 1.20. Here again, no significant change.

On the other hand, with virtual environments, nobody should be blocked with an old python version, so there is no strong argument to support them.

Given the problem encountered in #995, I'll open a PR restricting support to python >= 3.10. If there is a request to keep 3.9, we'll do it.

[oscarxblanco]

OK for me.