Job cleaner, use own thread pool for graph traversal

[albertz]

Via: #167 (comment)

[michelwi]

how much speedup do you see here?

could there be instances where the Cleaner threadpool is already too busy and using the original threadpool would be beneficial?

[albertz]

how much speedup do you see here?

I cannot really measure this because the NFS behaves different every time. But I think it reduced it.

could there be instances where the Cleaner threadpool is already too busy and using the original threadpool would be beneficial?

I don't exactly understand. You don't want that the cleaner is blocking the main threadpool. That's exactly what is fixed here. Otherwise you block the main thread (e.g. update_jobs) (see discussion in #167). You never want to block the main thread. (At least I don't.)