Lock not released on error; add error logging

[chris-branch]

I'm trying to track down the source of an error message that appears periodically in the logs for my Kong cluster (backed by Cassandra). The message looks like this:

connector.lua:269: [cassandra] failed to refresh cluster topology: timeout, context: ngx.timer

I believe this is coming from here:

https://github.com/Kong/kong/blob/master/kong/db/strategies/cassandra/connector.lua#L273

which ends up calling _Cluster:refresh() in lua-cassandra. While reviewing the code, I noticed that that _Cluster:refresh() obtains a shared lock, but the function has several exit points that do not release the lock. In particular, there are 6 places where an error return can occur where lock:unlock() will not be called. Example:

https://github.com/thibaultcha/lua-cassandra/blob/master/lib/resty/cassandra/cluster.lua#L583

The lock will auto-release after the 60-second timeout, but any callers may block until that happens.

Also, it would be helpful if _Cluster:refresh() could log a warning/error message for each of the early returns to help diagnose the root cause of any failures.