Reduce memory consumption of baseX

[jimmyli97]

Currently requires over 8 GB with a 1.6 GB database - I can't run it on a NIST computer

This can probably be fixed by outputting to a stream instead of holding all results in memory and then outputting at once.

[physikerwelt]

I guess most of the memory is consumed by basex itself. I think this is currently not high priority. For Wikipedia we will manage to fit it into memory on 16G instance. For ArXiv we use stratosphere anyhow.