Flatbuffers serialization could be faster

[djkoloski]

It looks like we could be carrying around scratch space and reusing it based on this issue: google/flatbuffers#6511

I think that's a little much to ask considering how much serialization code is already written for this benchmark, but it's worth keeping in mind.

[djkoloski]

Not interested in contorting the benchmarks to make one particular framework look better. These benchmarks should represent a reasonable amount of effort invested, but this seems unreasonable to me.