Implement basic queueing

[michaelsauter]

Pipeline runs belonging to one repository now run sequentially. If a
pipeline run cannot start immediately, it is created as "pending", and a
process is started to periodically check if it can start.

Since the pipeline manager service may be restarted, it check on boot if
there are any pending runs for the repositories under its control and
starts the periodic check for those.

We can improve on the design by adding a signal in the finish task of a
pipeline run that the run will soon finish, reducing the up to 30s wait
time for the next run.

Closes #394.

Tasks:

• Updated design documents in docs/design directory or not applicable
• Updated user-facing documentation in docs directory or not applicable
• Ran tests (e.g. make test) or not applicable
• Updated changelog or not applicable

[michaelsauter]

Please do not review this now. I would like to merge #480 first, then rebase this PR so that it becomes way smaller and easier to digest.

[michaelsauter]

@kuebler @henninggross @gerardcl ready for a first review pass. I'm not really happy with it but also unsure what exactly to improve. Let me know if you have ideas!

[kuebler]

LGTM! 🚀

[henninggross]

All in all it LGTM.
I am a fan of explicit variable names and would probably have called all variables concerning pipeline runs pipelineRun instead of pr (e.g. oldestPipelineRun instead of oldestPr), especially because outside of ods-pipeline the abbreviation PR in most cases means pull request, but that is personal preference and does not need to be implemented.

[michaelsauter]

@henninggross re: variable name - in Go, short names aren't seen as bad practice per se, rather it depends on the context. I kind of try to follow https://github.com/golang/go/wiki/CodeReviewComments#variable-names.