Make document structure processing parametrizable

[kermitt2]

Similarly as for the query attribute mention which can list different mention recognizers to be applied, we could introduce a structure attribute that will specified the (unique) underlying structure to be considered for processing a complete document (usually a PDF).

For instance default would mean no structure is considered (recognize and disambiguate entities in any tokens extracted from a PDF), scientific-article would mean to use GROBID model for scientific publication, e-book would mean to use GROBID model for structuring monograph document, patent-st36 a patent in XML ST-36 format, etc.

For each structure type - except the default one - some structured might be ignored because they are not textual or it does not make sense to apply a generic entity extraction, and some will be relevant. Ideally this could be specified in some config files.

[kermitt2]

structure query parameter is implemented see https://nerd.readthedocs.io/en/latest/restAPI.html#structure