Modularity restructure

[Archer6621]

This PR turns the chip demo into a fully modular framework. A non-exhaustive list of changes:

• Completely separate every module from one another, and collect them all in a modules folder:
  • Centralize the CI scripts into a single script that automagically runs for all the available module supporting tests.
  • Co-locate the relevant compose.yml bit with the module.
  • Add per-module configuration via a config.env file, exposing configuration values as environment variables.
• Add a new entry point script, chip.sh, which should be used to start/stop/clean/build the system or individual modules. Please view the new README for all the details regarding this script. It even support auto-complete for the module names!
• Add a new README with up-to-date information and quick start tips, and add a standardized README file for each of the modules.
• Standardize the routing between the modules, and add a layer of indirection over the core modules so that they can easily be swapped in/out by merely editing the core-modules.yaml file.
• Change the overall architecture such that the front-end no longer also sends to the response generator, but that it rather follows a linear route. This massively simplifies the architecture, without much of a downside (other than that you need to pass sentence data around through all the modules).
• Adds a properly setup and fully configurable LLM response-generator module using medalpaca 7b by default, which runs within a docker container and makes use of the GPU for inference.
• Also adds back some of the other alternative modules, such as the LLM triple-extractor, and the Gradio front-end, since they can be swapped in anyway via the core-modules.yaml if one wishes to check them out.
• Many other misc. changes and fixes.