Stage modularity, round 2

[mwhamgenomics]

At the moment, pipelines are split into stages but handling of intermediate files is messy and it's hard to modify pipelines. It would be better if:

• stages take generic parameters for input/output files, not predetermined file paths
• file paths are defined by the Pipeline object, or whatever the stages are being used by
• no wildcards - part from date-stamped files in BCBio, we should know what every file will be called
  • this might make output_files.yaml redundant
• checks for whether a stage should run uses presence of a reporting app stage is used as well as presence of files, not instead

We should also be able to mock a dataset, patch executor, run a pipeline and assert what all the bash commands were.

We should also make Stage objects as lightweight as possible, ideally removing their access to Dataset - see #395 for reasons why.

This might be a good opportunity to look at sciluigi