DCP stream names conflicting across different processes due to not seeding random number generator

I'm seeing a case where two Sync Gateway processes (running on different VMs) are trying to open DCP streams with the same name, and the root cause looks to be due to not seeding the random number generator in cbdatasource

This is causing the DCPReceiver to have it's OnError() called back, which in the Sync Gateway case causes the database to go offline and certain functional tests to fail, which were passing previously when using a TAP feed.

Sync Gateway 1 logs

Sync Gateway 1 pcap

Sync Gateway 2 logs

Sync Gateway 2 pcap

Note that in Sync Gateway 1 logs:

2017-05-03 17:20:59.351125 I | Opening DCP connection to cbdatasource-380704bb7b4d7c03

2017-05-03 17:21:00.402345 I | Opening DCP connection to cbdatasource-1408d2ac22c4d294

and in Sync Gateway 2 logs:

2017-05-03 17:21:01.114965 I | Opening DCP connection to cbdatasource-380704bb7b4d7c03

2017-05-03 17:21:01.759772 I | Opening DCP connection to cbdatasource-1408d2ac22c4d294

According to Dave Rigby: "it's certainly the case that a second DCP connection with the same key name in OPEN_CONNECTION will close the previous one".  

Original sync gateway ticket: https://github.com/couchbase/sync_gateway/issues/2514