Improve macro benchmarks data quality and coverage

[AdamGS]

Macro benchmarks currently focus only on query runtimes, in addition to that, we should track and measure:

1. Memory usage - could be either using DHAT or jemalloc
2. IO (latency and bytes read)
3. Separate hot and cold runs, purging OS caches in between queries and measure tracking each individually.
4. Hardware setup (machine, disk, os, etc) to make it easier to analyze historical data or compare the benchmarks between hardware setups
5. potentially perf counters? instruction counters?