Better scheduling and observability of the function image build process

[kwiatekus]

Description

Improve function build mechanism so that it builds function images according to the configuration:

• number of parallel builds
• computational resources for build runner
• target docker registry where the image should be pushed (changeable in the runtime, not at installation time)

Collect the following metrics:

• number of builds
• build error rate
• max queue size (peak value)
• average build time

Define default configuration per profile, ie:

• single build allowed at a time in evaluation profile. Additional build requests would need to be queued.
• 5 (?) simultaneous builds allowed in production. Anything beyond the limit goes into queue.

Build runner must not be a privileged container (see issue)

Reasons

Currently the build process execution time cannot be controlled. If 100 functions are pushed at once, all 100 build jobs will be scheduled simultaneously, trying to get computational resources.
In the future, we plan to have a central, multi-tenant build service. In order to evolve in this direction we need to introduce build scheduling already in single tenant mode inside kyma runtime.
Please keep the future vision in mind. Try to implement the build service so it would fit into multi-tenant mode.