Use marve along with(instead of) quantities

[Aazhar]

• investigate the options/profit of using marve for quantities parsing and recognition.

[kermitt2]

As marve uses grobid-quantities, the difference lies in the quantified substance/related information.
The problem with marve is that it uses Stanford Core NLP which is strict GPL (not even LGPL), so this is not really usable for us.
This is the reason why I used ClearNLP (now NLP4J), which is actually not worse than Core NLP for extracting dependency information.
The way ClearNLP is used in grobid-quantities is not so different than the way Core NLP is used in marve, it's just that it is very minimal in grobid-quantities because I think the right solution is to create a dedicated ML model where the syntactic/semantic dependencies provided by the common NLP toolkit would be used as feature, among others. But of course it's much more work :)