Naming convention of workflows

[oesteban]

Whereas interfaces can be tool-oriented, workflows should not be tool-oriented even if all the interfaces used by a workflow belong to a unique package (e.g. fsl).

I propose to change the workflows tree to a functional organization, where we have families of experiments (i.e. diffusion, fmri, morphometry) at the top level and then followed by the main processing blocks (i.e. preprocessing, reconstruction, model, reporting, etc.).

Interfaces that compose the workflow should be indicated in the documentation, maybe the scouting function that analyzes the workflows and calls the write_graph function can do this (print a summary with the node name and the interface module it belongs).

Looking at the current contents of the workflows module this should be rather effortless, although I must recon that I don't know any about fMRI workflows.

This discussion raises from #903.

[satra]

i like this layout. in addition every workflow should have a checker (enabled/disabled through the config) or explicitly called by the user to notify if any dependencies are missing on their system.

[satra]

also for the sake of backward compatibility we should import the workflows under the older path with a deprecation warning at least for the next two releases.

[oesteban]

Both requirements seem to me sensible and easy to implement 👍 . This will also ease a lot when merging code from other projects based on nipype.