Allow to retry operations (internally)

[danielmitterdorfer]

Currently, it is not possible to retry an operation on certain conditions (e.g. on failure or a network timeout). However, in some cases we might want to do this; not so much in regular benchmarks but rather when we execute an administrative command (like creating an index, checking for cluster health, ...).

In the first step, we will just provide this functionality internally in Rally as we slowly move towards a more generic ability to describe arbitrary workloads against Elasticsearch (i.e. this cannot be used in tracks yet).

In the next step, we will turn the currently implicitly defined operations at the start of a benchmark into explicit runners which can then be used as usual in a track. This will be done in #293.

As we gather more experience, we might also open up retrying to regular operations.