Upgrade to new version of lightning.

[xju2]

Move our package hadml to the src folder to prevent a mixture of locally installed and the dev version. Now you need to install the package to use it. The environment for testing the code is shown as follows:

.venv) [xju@login38] hadml >./scripts/check_env.py 
python interpreter: /pscratch/sd/x/xju/Herwig/hadml/.venv/bin/python
python version: 3.11.7 | packaged by conda-forge | (main, Dec 23 2023, 14:43:09) [GCC 12.3.0]
torch:  2.2.2+cu121
torch cuda:  True
torch cuda device count:  1
torch cuda device name:  NVIDIA A100-PCIE-40GB
torch cuda device capability:  (8, 0)
torch distributed     : True
lightning:  2.2.1
pyg:  2.5.2
torch_scatter:  2.1.2+pt22cu121
Test scatter_max in cuda.
out: tensor([[0, 0, 4, 3, 2, 0],
        [2, 4, 3, 0, 0, 0]], device='cuda:0')
argmax: tensor([[5, 5, 3, 4, 0, 1],
        [1, 4, 3, 5, 5, 5]], device='cuda:0')

I gave up the idea of conda environment and embraced the .venv, which can be easily created in VScode.

We may miss some exemplary input data for running the examples. We should put some small data into the data folder so that we can easily test if "old" training works.

[remilvus]

Maybe we should still have this command, so that it's obvious which python version should have all the correct versions of dependencies available?

[remilvus]

Can we change it to [800, 100, 100]? It's not very important, but this way it would be clear those aren't percentage points