Refactor interaction with Kubernetes

[blackandred]

The interaction with Kubernetes should be based on reconcilation loop.
There should be no API calls during the HTTP request, the underlying database adapter should use the reconcilation cache.

Advantages:

• Mitigation of DoS attacks on the API server and the application

Disadvantages:

• Probably two "threads", first for the HTTP application, the second for reconcilation loop, but it is not so obvious now, requires analysis

[blackandred]

There should be a separate controller that would communicate with Backup Repository via API internally.
For simple reason - controller can be scaled independently of Backup Repository due to different natures of those applications. Controller interacts just with API, while Backup Repository handles file uploads and needs to be handled like carefully like a stateful application.