Startup time measurement should wait until node start

[danielmitterdorfer]

Since https://github.com/elastic/rally/pull/719, Rally starts Elasticsearch nodes in daemonized mode (prior to that Elasticsearch was started as subprocess maintained by Rally). As soon as the Elasticsearch process has started, it will detach itself from its parent. As a consequence, startup time measurement is not very useful as it does not consider the node's bootstrapping phase. Instead, we should adapt an earlier approach to check in the node's log file when it has reached a defined state (i.e. the log line contains the "started" message).

Note that we are intentionally focussing on node startup time, not cluster startup time. The reason is that the preferred approach to start a cluster is to use Rally's start subcommand to start individual nodes and coordinate cluster start across multiple machines with configuration management tools like Ansible. Therefore, Rally has no notion of a cluster startup time as that part happens outside of Rally's control.