Arvados » Arvados Epics

Seems like this also needs to work for "bucket name = project UUID"? I imagine a case where a 3rd party service or scientific instrument is writing data to Arvados that consists of a number of different samples, in subdirectories, that belong in separate collections. We should ask potential users what they want to do with it.

I expect there are uses for "bucket name = project UUID" (and perhaps an "everything" bucket) but we might not want to jump in right away unless it's really needed right away. Semantics are a bit weird: creating an object called "foo" would never work, but creating "foo/bar" would (assuming "foo" is a collection in the project, or if we allow collections to be implicitly created that way). Implementing an efficient "list objects", iterating over all subprojects and collections with paging/cursors, might be an interesting challenge.