Use a shared Executor across all tasks.

[hamishmorgan]

Currently each task handles it's own parallelisation as best it can. This makes it somewhat problematic to set-up the processing environment, since each task must be configured independently. It would be more elegant if all tasks used a single Executor for their work. This would be also be an important stepping stone towards more generalised processing backends; for each to allow processing to be farmed out to a MapReduce infrastructure.