Profiles - Support different sets of rule weights depending on the project's use-case and maturity

[bvobart]

ML projects come in many shapes and sizes, this heterogeneous pluriformity also makes mllint an ambitious endeavour. Some rules are not equivalently applicable to every project, some rules may want to give different recommendations based on what kind of project they are looking at. Productionised or production-ready ML applications have higher requirements on SE quality standards than proof-of-concept projects. For example, proof-of-concept projects don't need to be entirely lint-warning free or have highly advanced deployment setups ready yet. Similarly, different tooling may be necessary / desired for a Tensorflow-project versus a PyTorch project versus an sklearn project.

mllint should therefore allow the user to specify a profile for the project in their mllint configuration. Such a profile will contain the maturity, i.e., current stage in the project's lifecycle (e.g. proof-of-concept, production-ready, production), as well as some way of defining the high-level kind of tooling that is used in the project (e.g. tensorflow, pytorch, <insert other ML project architecture inspired by some major ML library here>). The default would be a proof-of-concept project with auto-detected tooling.

The project's desired maturity level will be used to determine the weights of each rule. The auto-detectable high-level tooling architecture setting, can be used to steer recommendations towards or away from certain tools, e.g. TFDV instead of GreatExpectations for a tensorflow project.

Perhaps in the far future, this could also be used for linting ML projects in different languages such as R or Julia.