Make the OpenNeuro registration process a bit more intelligent for 'derivatives'

[prioux]

Right now, when a openneuro dataset is registered, we simply look at the top-level list of files and register those. It's fine most of the times, but some ON datasets come with a huge "derivatives" folder that thus gets registered as a single huge FileCollection.

I suggest the registration process look into the 'derivatives' folder and maybe partition it into smaller pieces, e.g. maybe one FileCollection per tool per tool's entry in there, or if the derivatives aren't already split by tool name, one per entry. That's probably a bit a fuzzy decision making given there's not a lot of standards about how derivatives are structured.