Process sparse fieldsets while generating REST-API response

The REST-API currently generates the full response and then processes the sparse fieldset conditions which return a subset of the fields of the JSON object. This is a simple approach but it's wasteful if only a small part of the result is needed. By pushing down the expressions that define which fields should be returned, only the needed parts are generated. This makes things like maxctrl list servers a lot more efficient if the statistics data that is spread across multiple threads isn't needed.

When combined with MXS-5220, this makes it possible to generate more efficient monitoring of MaxScales with MaxCtrl.