Maintenance for AsyncBulkCall

[maxfischer2781]

This PR fixes some internal issues of AsyncBulkCall. Major changes include:

• correctly call task_done for each item read from the queue
• maintain strong references for bulk tasks to avoid aborting them (see bpo#44665)
• clean up the dispatch task when there is no work to do

Closes #232.

---

Note: I considered cleaning up the dispatch task via a callback, the way bulk tasks are cleaned up. However, I could not find guarantees on promptness so had to assume using a callback may introduce a race between shutting down _bulk_dispatch and __call__ detecting that it needs to restart dispatch.

[maxfischer2781]

@giffels The newest flake8-bugbear release adds B019, which does not like our use of LRU caches on methods. Should I disable the check?

[mschnepf]

I have just one comment/question, But nothing critical. Thanks for your work 👍

[mschnepf]

Thanks for the change. Now it is less confusing. Merge it so 🖖

[mschnepf]

Looks fine. Thanks for your work.

[giffels]

@giffels The newest flake8-bugbear release adds B019, which does not like our use of LRU caches on methods. Should I disable the check?

Yes, please disable it.

[giffels]

Thanks a lot for your work. I will merge it once B019 is disabled.

[giffels]

Thanks a lot for your work! This PR is ready to meeeeeerrrrrrggggeeee! ;-)

[mschnepf]

Looks good 👍