drawingML shapes are independent, and having a separate parser for each
shape is a good thing, so that when we have an XML fragment like this:

<wps:wsp>
  <wps:txbx>
  ...
  </wps:txbx>
  <wps:bodyPr .../>
</wps:wsp>

where <wps:txbx> may contain nested <wps:wsp> tags, we apply the
properties specified in <wps:bodyPr> to the correct shape.

In contrast, doing the same for VML is not a good idea, see commit
476316bf (do reuse shape context,
2012-05-22).

So conditionally (for DML) maintain a stack of shape parsers, this way
the outer <wps:bodyPr> is parsed and the result is applied to the
correct shape, at the end resulting in correct inner margin for the outer
shape in the bugdoc.

This requires moving the context setup code from
the OOXMLFastContextHandlerShape ctor to setToken(), as only then we
know if this shape context will be used for VML or DML purposes. This
should be OK, given that
writerfilter::ooxml::OOXMLFastHelper::createAndSetParentAndDefine()
calls setToken() right after allocating the context.

Change-Id: I8c0d2f49cac76589fe269230698b203b5ca6996c
Reviewed-on: https://gerrit.libreoffice.org/39798Tested-by: Jenkins <ci@libreoffice.org>
Reviewed-by: Miklos Vajna <vmiklos@collabora.co.uk>