Draft project to hash out implicite pipeline with nipype.

If we are not interested in workflows or provenance tracking, the nipype engine is overkill and forces to use unPythonic style. However it brings us two nice features:

1. Not heaving to worry too much about filenames
2. No recomputations

To implement these, it requires explicit naming of each operation and writing the control flow as a data-flow program with tracked variable. These explicit steps are necessary to do the independence matching between the disk and the computation. Indeed, the disk-persistence has no notion of variables in and out of scope: everything is global. On the opposite, the computation is executed on the stack, and is thus fully defined by the values of the local variables.

Here we replace these by automatically generated names using hashs. This solution forgoes provenance tracking but opens the door to using imperative Python code to assemble the steps.