Correct management of model's input catalog and arg files

Redesign the way Experiments set input data to be passed to the Model's context. Should deal with all extreme cases:

- venv/conda execution: place catalog&args in model's input folder (cant be run parallel)
- venv/conda execution: place catalog&args in results/time_string/input/ (could be run in parallel)
- Docker: mount volume results/time_string/input/  <<>> /app/input  (can be run in parallel)