Improve cleanup after device disconnected event

Improve cleanup after device disconnected event

- Ensure that all contexts are removed from managers at once
  at end of disconnect chain right before unregistration from
  cluster singleton service
- Prevent closing some services multiple times during device
  disconnect chain
- Prevent some NPEs in DeviceInitializationUtils what was caused
  by connecting and disconnecting device with no time in between
- Unify all openflow managers to implement OFPManager interface
- Ensure that when we will fail to set device to SLAVE we will wait
  until device is removed from operational DS and then continue

Resolves: bug 6672
See also: bug 6710, bug 5271

Actual work done by Tomas Slusny. Colin Dixon merely fixed a merge
conflict.

Change-Id: I3845c7d2a9147125b185d3abcc3fc8dd63ba5da8
Depends-On: I15f89cbd43310b0a8e4b7ac78595eeaaa82a65ee
Co-author: Tomas Slusny <tomas.slusny@pantheon.sk>
Signed-off-by: Tomas Slusny <tomas.slusny@pantheon.sk>
Signed-off-by: Colin Dixon <colin@colindixon.com>
(cherry picked from commit 4a89e389589676a2b6975096ba721cd4c354e1bd)