add json outputter for debugging

[danpf]

I think this is reasonable to have.... I've had some interest myself in better mmtf 'diff' tools, and had a complaint or two from rosetta users about diffing abilities.

let me know what you think!

i can add an app to do this too if that's of interest

[speleo3]

insCodeList and altLocList strings get dumped as null bytes: "\u0000". Would look better as empty string, right?

[gtauriello]

Some comments and observations:

• It doesn't seem to be necessary to convert int8 to int32. Actually (due to a typo I guess), the bondResonanceList field is not converted for the json output. On my machine I could safely drop those conversions for bondOrderList, bondResonanceList and secStructList. Any reason to keep them?
• The extra properties cannot be dumped into valid JSON if they contain encoded data. I tested it with the attached MMTF file (which is the file generated in the "extra data fields" test case of mmtf_tests.cpp). At least Python and Sublime complained about the generated JSON file. So we should either drop the json-dump for the extra properties or just dump the keys.
• The traverse.cpp example code had a json dumper which I quickly extended to compare with the to_json function. The advantage in traverse.cpp is that it uses null for data which wasn't set. This can be confusing otherwise for optional numeric fields (e.g. resolution) which are set to std::numeric_limits<T>::max() if they are not provided in the input MMTF file. Also traverse.cpp does some custom pretty-printing suited for MMTF-content. On the down-side, the traverse.cpp is an ugly C-style C++ code (my apologies since I wrote it) which was originally meant to allow diffing with the mmtf-c library and grew beyond sustainability.
• I also included the to_json function in traverse.cpp to be able to call it from there.
• I don't think that the to_json.cpp app is useful within tests and I would drop it. The code is compiled anyway, so a pure "it compiles" test is not adding anything. It could be useful in examples but the traverse.cpp could cover that (btw: you can also call the StructureData::print functionality from traverse.cpp).

So to sum up: If we have such a to_json function as part of the library it should handle unset optional fields. I suggest to drop them since this already happens within GroupType objects (handled automatically via msgpack_encoders.hpp). Also the items above (unneeded conversions, faulty JSON with extra props, unneeded to_json.cpp) should be fixed.