Refactor CI stages

Currently, our CI tasks are split into three stages:

• build
• test
• deploy

The test stage contains only one task: system_tests.

The CI stages work so that the previous stage must finish before the next starts. The build stage contains two heavy tasks: build_deb and build_rpms. They take a lot of time. Additionally, they produce significant artifacts - binary packages. We limit the lifetime of these artifacts to 5 minutes to limit disk space usage.

It has one unexpected consequence. The system tests may be run only 5 minutes after finishing the build stage. It causes the repeating of the system tests pipeline to be impossible.

I propose to refactor the stages. We should split the heavy tasks (system tests, build packages) from the fast, light tasks:

• check-in
• heavy
• deploy

It allows us not to start time- and resource-consuming tasks when the basic checks fail. Additionally, it fixes a problem with creating the system tests pipeline.