Allow the setting of the a pg_rewind interval and timeout so that the
retry behaviour of pg_rewind can be customised. The defualt values if
these are not set in the clusterspec are 0s which results in the
current default behaviour of only trying to pg_rewind once if
usePgrewind is set to true.

GC has implemented cascading replication which requires the retrying of
pg_rewind to avoid expensive pg_basebackups.

With cascading replication, the old sync is rebooted to repoint the
recovery at the new primary. The old master will try resyncing against
the new sync, while it's rebooting, which causes resync to fail. This
falls back to pg_basebackup meaning the old master doesn't recover
for several hours.

Make the forcing of a checkpoint on the primary before performing a
pg_rewind a cluster spec option so that it can be turned on or off.

Add pgrewindInternal, prgrewindTimeout, and pgrewindCheckpoint to the
cluster spec documentation.