JSON files report both CPU time and real time

Each benchmark has a preference for real time or CPU time, depending on what's being tested, but gbenchmark always stores both values in JSON output files. This means it's not possible to programmatically determine, when reading a JSON file, whether CPU time or real time is relevant in a particular benchmark.

It seems that when the benchmark declares a preference for real time, I could parse the benchmark name for the string `/real_time`. But that seems hackish...