Implement data management API for worker nodes

[DanielVoogsgerd]

Moving data into a worker node is often quite a hassle. We have to copy data into the workers data directory, after which we have to change the absolute path to match the new location.

I suggest the following changes:

• Documentation should reflect the difference between Assets and Datasets.
• Assets must use relative paths relative to the data.yml, in order to be portable.
• a brane data push or brane data upload needs to be implemented, which takes an asset. Which sends it to the worker node.

@lut99, what do you think?

[Lut99]

Good point!

1. Yes plz, even I'm always hazy on which does which;
2. That's OK, provided that nothing in the code directly uses assets anymore OR any part that uses it has its code updated to canonicalize the path before mounting it to a Docker volume if it doesn't already (Docker requires absolute paths).
3. I'm against a brane data push; it's not the scientist's job to make data available on the worker node simply because it requires updating the policy on the worker. Instead, we can make a branectl data push (which is the worker-side tool, after all). Longer-term, I have in my mind to either have central host a package-like repository for data to which scientists can push (but necessarily isn't shielded by policy); or to make an optional flag for workers to allow "guest data".

That aside, I agree it's a hassle. Something like branectl data push is most definitely a very nice option (or perhaps branectl data build, to mirror brane's operation but for worker managers).