Improve Dockerisation of Forecasting Agent

[markushofmeister]

Enhance current dockerisation / stack deployment

• [x] Include setup.py and requirements.txt with pinned version numbers for Docker images
• [ ] Update Dockerfile: Include StackClients only in stack image (condition within Dockerfile vs. multiple Dockerfiles)
• [ ] Investigate what's driving the massive size of docker image, and try to remove unnecessary dependencies
• [x] (Automatically) publish Docker image upon building: release script vs. publish via json config files for stack-manager
• [x] Include debug configuration for debugging when spun up via stack-manager
• [x] Potentially remove Stack name from docker-compose files

[markushofmeister]

Main driver behind massive docker image are nvidia cuda torch dependencies

2.6G    /usr/local/lib/python3.9/site-packages/nvidia
1.3G    /usr/local/lib/python3.9/site-packages/torch
1.2G    /usr/local/lib/python3.9/site-packages/torch/lib       
861M    /usr/local/lib/python3.9/site-packages/nvidia/cudnn    
860M    /usr/local/lib/python3.9/site-packages/nvidia/cudnn/lib