HierarchicalStreamReader should be able to lazily read

Currently all implementations of HierarchicalStreamReader) have read the entire stream up front before being passed to the converters. This is because of a flaw in the design of the interface.

This causes a lot of unnecessary overhead (both in memory and performance) and XStream starts to chug when deserializing larger documents.

The interface should be modified so the underlying stream can be read incrementally, converting to objects as it goes.