MixFunn cluster experiments

Place all files in a project directory on a Slurm cluster. Install requirements.txt
in your Python environment. Edit PYTHON_BIN in run_training.sh to its absolute
Python path (or export PYTHON_BIN before sbatch). Submit from the project directory
or edit PROJECT_DIR to the absolute path containing train_experiment.py.

Download experiment.json from the website and run:
  sbatch run_training.sh experiment.json

After completion open runs/slurm_JOBID/curves.json in the Cluster experiments tab.
The website currently uses manual submission and local results import.

The default example uses no supervised QuTiP loss. QuTiP remains a dependency
for exact evaluation. Model architecture, loss weights, time grids and FNO
settings are specified in experiment.json; the runner validates the configuration.
