Move branched db detection to the sentinel

[sgotti]

Currently it's the keeper that detects if a db is branched from its master. We want to change this:

• move the detection to the sentinel
• have the sentinel, when it detects a branched db, remove the bad db from the clusterdata and create a new db with a db InitMode of type resync. The keeper will converge to the sentinel required state (and it'll continue to use pg_rewind when possible and enabled).

This has different advantages:

• The keeper will resync only when asked for and not when it wants
• since it's the sentinel detecting bad dbs it can better control standby removal from the cluterdata and from future proxing (Allow clients to connect to standby replicas in RO mode #132)
• The resync db initMode can also be used for initializing a standby cluster Stolon "standby" cluster #157.

[sgotti]

Implemented in #218.