Use `uuid` in dataset dependencies

[ilongin]

Currently we use combinations of dataset name + version, but this is not stable i.e. someone could re-create dataset and it's name + version will be the same, but content (actual dataset rows) may not. Now that we have uuid inside dataset versions, we should use that instead. This will make code simpler and schema smaller as well.

[shcheklein]

Currently we use combinations of dataset name + version, but this is not stable

so, do we? or do we rely on id after all?