Support for collecting outputs from tools that produce arbitrary number of outputs at runtime about bioblend

[jmchilton]

@unode One substantial complication is that <discover_datasets> actually is tweaking existing logic that applies to every output of every tool. So any tool could be a A tool - not even Galaxy knows which type of tool it is until the job is complete.

In the test framework instead of waiting on the history or outputs, we wait on the job (thanks to @davebx) - this makes sure we collect all the outputs and avoid race conditions. I have not touched the objects code - but this would be my recommendation for what to do there as well.

[nsoranzo]

@jmchilton Tool.run() in BioBlend.objects presently returns a list of output HistoryDatasetAssociation objects created from rval['outputs'] of POST /api/tools. This should also probably be updated, if we move to waiting on the job. Suggestions?

[jmchilton]

Maybe use

        * GET /api/jobs/{job_id}/outputs
            returns output datasets created by job

once the job is complete. Looks like there is not yet a bioblend entry point for that though - https://github.com/galaxyproject/bioblend/blob/master/bioblend/galaxy/jobs/__init__.py.

[nsoranzo]

There is JobsClient.show_job(job_id)['outputs'] which returns a dict instead of a list, but I don't think that's a problem.