If ctx->expandEntities is false, then don't expand entities in data.
Patch from Bob Dunlop, <rdunlop@guralp.com>.
I'm not 100% sure of the correctness of the behaviour of entity expansion in
general -- shouldn't expandEntities toggle between either doing an entity
content lookup and substituting the result in the data, or causing a specific
"there's an entity here" callback? At present it seems to toggle between
lookup/substitute and treating entities (including their leading ampersands) as
character data, which seems wrong. To be reviewed.
marker, leaving us in stream level after <![RESTART[, which meant that the closing bit of the marker
was interpreted as stream-level content. Also, the state machine now handles restart markers at high
level as well as low level; this simply means that a (false) error is not issued in the case of
perfectly well-formed Streamed XML containing restart markers.