Refinery custom docker parent image

[FelixKirsch]

Is your feature request related to a problem? Please describe.
Some refinery services depend on storage-intensive python packages like torch, spacy, sklearn, etc. These packages are installed again in every service. This increases download and build time.

Describe the solution you'd like
Defining specific docker parent images could prevent this behavior. These parent images would be shared between the services and therefore no duplicate installation of the packages would be needed. Docker shares the parent layer between different images so that disk space is only required once.

Describe alternatives you've considered
Keep the current functionality.

[FelixKirsch]

• define requirements files for the different parent images (2h)
  • How many are needed and how specifically do they need to be?
• create a Dockerfile for every parent image (1h)
• use pytorch cpu build as standard for the pytorch packages (1h)
• create drone.yml file (1h)
• create manifest for docker multiarch (1h)
• change the parent image in every service, and change the requirements (8h)
• testing (2h)
• review (2h)

OVERALL 18h