Separate track specification from track execution

[danielmitterdorfer]

We should not put tracks (i.e. the actual benchmark specification) directly into core but should provide some kind of benchmark repository (think Maven central but for benchmarks). However we still must be able to develop benchmarks locally.

We should also introduce a logical URL schema for track data URLs (e.g. $ROOT_URL/$benchmark-name/$index-name/$type-name/)

[danielmitterdorfer]

I thought about it and think it would be best to do this as late as possible. The API around tracks is pretty much incomplete and unstable. By keeping everything in Rally for now we have more control over which benchmarks exist and can easier refactor them.

[danielmitterdorfer]

A minor detail: We should also move track.py to top-level when the split is done to simplify import statements a bit (import rally.track instead of import rally.track.track).