Do not keep all in memory part1

Created by: giuseppe

some work in progress of not keeping all in memory.

This is basically a limit of 10M for the stuff kept in memory and the rest is streamed directly to disk.

It is not complete as download_part doesn't honor this.

The plan is that the different parsers will, with time, be able to parse the data on the fly without buffering the entire document. For now, just limit them to use the first 10M