strict_dns: allow to keep stale targets for a specified amount of time

Problem:
When a DNS does not return all targets on each resolve, `strict_dns` will continually be adding and draining targets. For this case `logical_dns` is better, but it's still not great, it keeps track of a single logical target even though connections are made to several, for the given resolution interval all connections and traffic will flow to a single IP address, also I really don't know how outlier_detection and other advanced features can play with a `logical_dns` cluster.
`original_dst` cluster supports cleaning unused targets and be an option, but it's also probably unsafe since it allows connecting to any IP address.

Proposed Solutions:
- Adding support for `cleanup_interval` for `strict_dns` clusters like `original_dst` does, and remove them when they have not been returned for a given time.
- Adding a "mixed dns" cluster which supports this sort of behavior.