Tune up backend GHA

[sergejparity]

Refactored workflow backend.yml by splitting it in four parts. Each job into its independent backend_fmt.yml, backend_check.yml, backend_docs.yml and backend_tests.yml respectively.

The goal of this change is to overcome limitations of the Swatinem/rust-cache GitHub Action. When multiple jobs within one workflow uses above mentioned GHA and executed in parallel it often fails to restore rust cache. Possible two solutions, make jobs using rust-cache to run sequentially within single workflow or setup separate workflows for every job. The latter approach is more reliable as allows to execute jobs in parallel despite little overhead of spinning up additional runners.

As a result workflows execution time reduced almost 3 times. From 15-18 minutes to 5-7 minutes.

[jsdw]

I'm a bit sad that we need to split into multiple files to make thie work better (though well done on reducing the build times so much; that's great :)).

Could you achieve a similar thing by using one file but tagging the rust cache invocation in each job with some different thing to give each a unique cache or something?

Still; I'm ok with it either way; just wondering really because I could see this sort of change being applied across a lot of projects.

[sergejparity]

Could you achieve a similar thing by using one file but tagging the rust cache invocation in each job with some different thing to give each a unique cache or something?

I've tried that approach as well, but looks like when it happens within one runner those caches corrupting each other. And it always a lottery whether it will work or not. Probably because it populating cache to $workspace/target and in this case we have one workspace - backend. We could checkout repository 3 times into different subfolders, but this would be even more ridiculous setup.

So either in sequence or in separate runners.