Collection Models and State Revision

[jmchilton]

A 2018 revision of collection models and job state are probably in order along the lines of https://github.com/galaxyproject/galaxy/pull/5013.

In particular I think we need to:

• Stop associating every input collection dataset with the jobs it is being used to create - we are tracking collection input associations and that should be enough in theory hopefully. Just associate the collection. What we are doing right now just doesn't scale well. This won't be easy because job polling depends on these datasets being present.
• Fix on_string for collection jobs at the same time. on_string is ridiculous for tools with collection inputs because these datasets are picked up as inputs and so random collection contents HIDs are used to give names to collections.

This also would bring us a little closer to allowing dataset collection contents be something that isn't an HDA (https://github.com/galaxyproject/galaxy/issues/1810).

[jmchilton]

Note to self - here is optimization WIP that didn't help because of this problem https://github.com/jmchilton/galaxy/tree/1805_tools_ready_fix.