Changes

ZIP: Streaming option to defer tree walking.

By default, the ZIP export collects the metadata of all objects it
intends to export before actually starting to output something. This is
sensible when there is a link structure between the various objects that
needs to be rewritten, since we need to know an object's metadata in
order to calculate its filename.  The disadvantage of this method is
that for large sets of data, it may take quite a while until we deliver
our first bytes, which may cause the client or some intermediate proxy
to timeout before we start delivering.

This commit introduces the '&stream=true' parameter to the ZIP export.
When given, it causes exporting to start live while we walk the
aggregation tree: Non-aggregation objects are written as soon as they
are encountered, aggregation representations as soon as all their
aggregated items have been written. This can reduce the delay
drastically. However, the link rewriting table used for each object
includes only those items that have already been seen, thus some
textgrid: links might remain unresolved in the exported files.

Added a search client implementation with a Collection interface

Refactored rootObjects to be an iterable

Removed unneccessary expansions of all root objects

Fixed NPE when AggregationTree is used for non-aggregations

ZIP: Added 'filter' argument to search export