[helm_resources_manager] use cache instead of a plain client

[miklezzzz]

Overview

Currently, helm resources manager creates a resources monitor for every enabled module that has a helm release. In its turn, a resource monitor periodically checks whether all the resource of the corresponding helm release are presented in the cluster. These checks are performed by retrieving partial metadata for every object of the release via List operation. The main gist of the PR is to get objects' metadata not from the cluster API directly, but rather from the cache, decreasing amount of List operations and speeding up some of the operations related to helm releases (checking if a helm release should be applied, running periodic checks for absent resources).

What this PR does / why we need it

• Helm resources manager gets a cache and shares it with resource monitors it creates.
• The cache has labelSelector filter that should help cache only related objects (objects, deployed by the addon-operator).
• In order to tell the difference between related and not related objects, a post-renderer was introduced.
• The post-renderer applies an additional label (heritage=addon-operator by default, yet configurable) to every resource the operator deploys (see helm post-renderer functionality).
• The post-renderer is supported by both helm3 and helm3lib implementations (with quite noticeable discrepancies in realizations, so using helm3 is discouraged due to operational overhead).

Special notes for your reviewer