Performance benchmarking improvements

As discussed offline, we think there are a few improvements we could make in how we do performance benchmarks for H3:

• use some standard benchmarking framework instead of rolling our own (which is what we currently do)
  • a framework could provide nice features like warm-ups, advanced stats (min vs avg), etc.
  • what options currently exist for C?
• standardize the benchmark format to make it more machine-readable (like in Go), which would make it easier to automatically compare timings between diffs
• avoid doing benchmarks on laptops and/or finding a dedicated benchmarking machine
• random (but deterministic) inputs for improved input coverage
• benchmark performance tracking/plotting over the timeline of diffs, like what is done in Python using asv for libraries like numpy

For context, this issue came up after seeing a lot of variance in the benchmarks I was running on my laptop for https://github.com/uber/h3/pull/496

Start by locating H3's current performance benchmark implementation, since the issue does not name files or tests. Review C benchmarking frameworks and the linked Go benchmark format, then compare the possible scope with Python's asv approach. Done would require an agreed design for benchmark execution, machine-readable results, deterministic inputs, and performance tracking.