Apply size limit to node structures.

[danielballan]

Unlike array and dataframe we do not generally know the byte size up front. We work this in two ways:

1. Try an optional est_nbytes() method that Adapters MAY implement to declare their (approximate) size up front.
2. Within the node serializers, track the cumulative size of data exporter and give up if the limit is reached, raising a specific exception type with meaning to the server.

The UI can then be updated to support node export. It should handle this error correctly.

[danielballan]

Latest thinking on this: before handing off to the serializer, walk the node just to tally the size of its children.