Upgrade NumPy to >=2.1.0

[AlexKurek]

Feature description
For latest NumPy (2.2.2):

Compare to the upper plot here: #222 (comment).
NumPy (2.2.2) is faster, in some cases >2x.
Upgrading NumPy to >=2.2.0 seems to be the only thing worth doing for NumPy at the moment.
Code:

### no MKL, latest NumPy
python -m venv nomkl_latestnumpy
source nomkl_latestnumpy/bin/activate

pip install numpy
pip show numpy
pip install setuptools PyQt5 pygount scipy

cd npbench/
python run_framework.py -f numpy

pip install pandas matplotlib
python plot_results.py

deactivate
cd ..

Related: #240.

[AlexKurek]

Currently trying to build Fedora 41 crashes on Numba:

Collecting numba<0.61,>=0.60 (from codex-africanus[dask]->-r /opt/lofar/requirements.txt (line 21))
[...]
RuntimeError: Cannot install on Python version 3.13.1; only versions >=3.9,<3.13 are supported.

Either way, codex-africanus does built manually in Python 3.13. Tagging @sjperkins.

[tikk3r]

Numpy 2 is a major pain, so I'm not eager to update just yet. It can be reopened once the ecosystem stabilised around it a bit more.

Also, please be mindful about whether tagging developers of packages is really necessary. The installation problem doesn't seem relevant to this issue (numpy 2 and Python 3.13 are two separate things to tackle), but also these containers are not a hard reference for stuff working or not working and are not affiliated to any of the packages included. I don't want this repository to become a source of noise for them. Issues are best raised on their own repositories and with the respective package's recommended method of installation.

[AlexKurek]

Just to show you for the future - Numpy 2.2.2 compiled with mkl is even faster than without it:

[AlexKurek]

I tried using Python 3.13.1 with GIL enabled. Its more or less the same as on the plot above (Python 12.7).
EDIT:
Building using Python 3.13.1 with GIL disabled crashes.