Allow targeted benchmarks of a change against a set of predefined repos

When doing performance-specific changes in Gitaly then the result can often be that the change improves performance for one type of repositories while it regresses performance for others. This is e.g. the case for pooled/non-pooled repositories or repositories that are huge in different ways (e.g. an unreasonable amount of refs, commits, size or whatever).

Benchmarks run manually only go so far and don't really provide complete coverage. We should define a set of repositories with different properties and allow to run targeted benchmarks against all of them.