Delete from L1 on writes when in an unknown state

[ScottMansfield]

This issue is directed at maintaining consistency between the L1 and L2 caches behind Rend.

There are some cases in the request orchestration (orca package) where a write has successfully been made to the L2 cache but L1 returns an error and is thus in an unknown state. In order to preserve consistency between the two levels, a delete should be sent to L1 when it's state is unknown.

In the case of an IO error, it is assumed that the memcached process has been killed or is otherwise in a bad state, so those errors will not be captured by this logic.

[ScottMansfield]

As a part of this, any errors that are encountered that cause Rend to delete from L1 should not be returned to the requester. The intention of this issue is to mask any temporary errors given by Memcached.