Fixing TTL in Cassandra

[subnetmarco]

Fixes the TTL problem mentioned by @chrisanderton in #1054.

Also introduces the ttl_on_failure property in the configuration file, because 3600 may not be an appropriate value for every environment.

This PR is to be merged in both next and release/0.8.0.

[chrisanderton]

@thefosk this will only fix half of the TTL issue on Postgres - the TTL is now being correctly passed in - and i like that you've refactored it to config rather than seeing a random 7200 in the code :)

But.. when using Postgres the nodes will still be removed from the cluster after a now default 3600 seconds - the TTL will never move because it is calculated based on the created timestamp of the node plus the TTL (created at will never change):

https://github.com/Mashape/kong/blob/fix/ttl/kong/dao/postgres_db.lua#L230

I think there are two options:

1. Update the TTL based on the expiry of the previous TTL - i.e. existing TTL expiry + 3600 - but this might need an additional query which seems wasteful
2. Update the TTL using Now + TTL - something like local expire_at = now + (ttl * 1000)

[subnetmarco]

@chrisanderton, the last commit should also cover that use-case.

[coveralls]

Coverage decreased (-0.2%) to 59.231% when pulling 1f8b9c1 on fix/ttl into c08b22a on release/0.8.0.