When we requeue a job, we need to process subsequent skipped jobs,
moreover, we need to do this in a specific order.
Previously, we had an order by stage_idx but after introducing
same-stage jobs, this disappeared. We recently fixed this problem
for stage-based approach. However, the problem still exists for
the same-stage pipelines.

In this commit, we are ordering jobs by stage first,
then their DAG relationships.

These changes are behind a FF ci_fix_order_of_subsequent_jobs