The problem was that SwXTextEmbeddedObject::getEmbeddedObject() returned
an empty reference for those embedded objects, so the HTML filter
couldn't extract their content when it wanted to do so.

It turns out the reason for this was that the DOCX importer only handled
the replacement image + raw native data for the object. Fix this by
creating the embedded object with the correct CLSID and import the
raw data into the empty embedded document model.

This is similar to what is done for XLSX-in-PPTX in
oox::drawingml::ShapeExport::WriteOLE2Shape(), just for the import part.

Change-Id: Ieb1dcb1774d2d4da00117e3a35160053066c78aa