backend: move initial createrepo check from dispatcher to worker
    
    Previously we faced concurrency issues between post-build copr-repo runs
    and these pre-build copr-repo runs.  The pre-build copr-repo could hit
    the small time window when 'repodata' don't exist when craterepo_c is
    run (moving .repodata to repodata) and start running our own 'copr-repo'
    process mistakenly.  That would be a long-running process (full
    createrepo_c run, without --recycle-pkglist), and more, would be blocked
    by other concurrent copr-repo processes (so it would be eventually
    killed by timeout -> the build would continue -> and it would fail
    anyways - regardless of the actual mock results, issue #1222).  This
    race should be closed now.
    
    Also, stop running copr-repo before build is started (again), and "only"
    wait till the repodata become available (and fail if they won't, after
    certain time period).  We used to have problems with this approach
    before -- but we did not fail the builds, but rather kept waiting
    indefinitely.  Neither approach is ideal, but with the current approach
    we should better motivate people to report issues.
    
    Also move the logic out from BuildDispatcher to Worker so we never block
    other builds by those checks.
    
    Fixes: #1222