descriptor limits to account for bigcouch sync spikes (#4935)

Change-Id: I242fba31f961b6139ec641e1708b170f5c0d009b

In a multi-node couch deployment, it was observed that the Service['stunnel']
would be activated, and then later a stunnel::client was created which would
trigger an Exec['refresh_stunnel']. Because of this, and the ordering hints
that were in place, the service would get started, and then the couchdb
databases, users, designs, etc. were being put into place and then a stunnel
client was created, triggering the refresh_stunnel exec, which would cause
an interruption in the connectivity and result in failures.

This change replaces the Service['stunnel'] hint with the the
Exec['refresh_stunnel'] to make sure that the stunnels are fully setup before
attempting couch operations.

Change-Id: I33ddd24884b3c23a1df5555ca53ca65cd703da50

Change-Id: I318944a6872a53ff9c533704514da339426d9401

Site_couchdb::Bigcouch::Settle_cluster/Exec[wait_for_couch_nodes] waits
60s for all nodes to be member of the cluster. Because we deploy to
multiple nodes in parallel, not all nodes are ready at the same time,
so we increased the timeout from 60s to 120s.

This only works with the latest patch to puppet_couchdb