Clean up stale Git processes when Gitaly exits
As part of an incident we have been derailed by a Git command which has been hanging for 1,5 months without doing anything. The process was stuck in its signal handler due to a deadlock. We should investigate whether we can improve the upgrade logic to properly kill off all processes which shouldn't survive between the different Gitaly instances. We may want to consider use of CGroups for this. Furthermore, we may have to use something else than SIGTERM given that SIGTERM wasn't able to kill the process.