Implement benchmark for get_json_object

[ttnghia]

This adds a benchmark for get_json_object which allows testing performance of the function with input having various nested levels.

Also fixes some compilation issues in the benchmark code.

---

Build benchmark

The build command is similar to this:

./build/build-in-docker clean install -DGPU_ARCHS=75 -DCPP_PARALLEL_LEVEL=16 -Dsubmodule.check.skip=true -DskipTests -DBUILD_BENCHMARKS=ON -DBUILD_TESTS=ON

Run benchmarks and compare

Run this before and after each code change:

./benchmarks/GET_JSON_OBJECT_BENCH --json <before|after>.json

Then compare (using nvbench_compare.py):

 python nvbench_compare.py before.json after.json

[ttnghia]

build

[ttnghia]

build

[res-life]

I compiled and tested sucessfully.
LGTM