io_queue: Destroy priority class data with scheduling group

[xemul]

When a scheduling group is destroyed, the IO queue's priority_class_data
that corresponds to it stays intact. It's not actually a leak, as when
the reactor is destroyed all class data objects are collected.

However, if another scheduling group under the same ID is created, the
IO queue will re-use the dangling priority class for the newly created
group, but its name and shares won't match the new group until rename
or shares update. And it's not nice that IO class remains alive anyway.

Since scheduling group destruction must happen after all tasks from it
are completed and picked up, the class destruction code may also assume
that no queued or in-flight requests exist.

Test included.

[xemul]

upd:

• assume that when priority class is destroyed, no queued or executing requests exist

[xemul]

@avikivity , please review

[avikivity]

I unmerged it, it needs rebasing after the data_sink API changes.

[xemul]

I don't see how it conflicts with data_sink API changes. Maybe it's API changes that need to be unmerged, because they clashed with #3018 's on counting_data_sink? OTOH, API level 8 in "library mode", i.e. without tests, demos and apps, still compiles fine

[xemul]

Created #3040