Add support for running destructive and long running tasks from
MANAGED state, such as configuring RAID or doing burn in.
The changes since Kilo is the addition of caching of clean/zap
steps to make the get_clean_steps or get_zap_steps API faster.
Also adds support for those APIs to wait for interfaces that must
take asynchronous steps to get the steps (like IPA).
This will require allowing APIs to return an empty response
and header that indicates when to check back.
blueprint: implement-zapping-states
Previously-approved: Kilo
Change-Id: I09491472d713fd7930f5592ae168cad3f92a16c2
c0ae0b6a63