[EDP] Refactor job manager to support multiple implementations

The original job manager was written to execute and monitor jobs via an Oozie server. As we add new provisioning plugin types, for example the Spark plugin, we will need new EDP implementations for clusters that do not support Oozie.

There are 3 basic operations that the job manager currently performs:
* run a job
* get job status
* cancel a job

If the job manager can select a class which implements these three operations for a particular cluster, then the job manager can be generalized and deal only with high level logic and update of job execution objects.

(detailed spec on its way)