Test cases: File hashing and nested state

[effigies]

In today's call, there was general agreement that we will be accepting fileformats as a Pydra dependency. The justification is that the strength of Nipype is in combining file-oriented shell commands with Python glue code, so a full-featured file type library seems reasonable.

A significant constraint for implementing generic hashing (#644) is nested data types. For the most part, we are content to leave the problems of deep nesting to downstream tools that want to use it, and __bytes_repr__ gives them full flexibility. One particular case that we must be very careful of is nested lists of files. Probably lists-of-lists of files are the most complex a single tool is likely to take (consider pairs of pre- and post- masks for each registration stage in antsRegistration, or a list of (effect, variance) pairs of files in second-level GLMs).

We do also need to consider the impact of state splitting and combining. Here, we might imagine splitting across subject, session and run. We should write tests that ensure that splitting and combining state does not affect whether a file is or isn't coerced.

In comments, we can workshop some specific tests, but just posting this now before I forget.