We want to capture metrics related to the availability impact of replicated index builds, specifically the problem described by SERVER-112315.

We should expose two serverStatus metrics:

• The last amount of spent between receiving a commitIndexBuild oplog entry and an index build being committed. This should be captured in any replication state (either secondary or startup recovery).
• The last amount of time spent between voting for commit (as self) and actually committing.

This will tell us how long a node was blocking replication by building indexes.