Test Case for Extended Metadata Handling with Pulsar

[jmchilton]

I failed repeatedly at this task in late 2019 and early 2020, my latest failed attempt is at #9227 I think.

Getting some consistent stack traces so we can figure out what potential problems there are with running Pulsar with extended metadata on would be awesome and allow rapid progress.

In addition to being more efficient in these scenarios with remote storage and distributed compute, running Pulsar in extended metadata would solve many classes of output collection errors. Stuff like discovery of datasets (galaxyproject/pulsar#212 and galaxyproject/pulsar#239) would happen on the compute node and we wouldn't need to worry about transferring the right files back to Galaxy.