Add modularised pipeline methods

In the O3a implementation all jobs were assumed to be LALInference jobs, but this was inflexible. We should allow any pipeline to present itself to an asimov bot through a straight-forward API. This will allow for additional monitoring applications to be folded-in as well.