qa: add osd thrashing tasks
set up a collection that thrash the osds. maybe 3 facets:
- a bunch of cluster definitions (3, 8, 16 nodes?)
- a few different thrasher settings, with various tunables.. super frequent thrashing, less frequent (so that recovery has a chance to kick in for each change), etc.
- a bunch of workloads. things like iozone, radosbench, bonnie that generate lots of io.
currently the yaml has to compose cleanly, and both the thrasher and workunits are in the tasks section. need #1497