Stop saving prepared datasets in outputs

[andrea-gariboldi]

This PR will:

• Stop saving the prepared datasets in the output of each run;
• Save dataset metadata in a separate json in the /shared dir in the output folder of a run;
• Use the dataset metadata for forking (for non-forking runs, default behaviour).
• Update docs;
• Add tests;

This change affects both the host launcher and the container runtime. After rebuilding the dev image as per the documentation,

PYTHONPATH="$PWD/src" python -m agentomics.cli.run ....

Instead of using the agentomics-run command.