Use ujson to parse json files

[KEggensperger]

This should give us some speedup when reading large data for tabular benchmarks. I would propose to make it an optional dependency, i.e. try importing it and if it fails fall back to normal json.