consider using raw measurement types in all applicable PKB-YCSB benchmarks

YCSB has a number of "measurement" classes, all of them are only capable of reporting latency at granularity of 1ms except for the RAW measurement class which reports latency in microseconds.

For a number of NoSQL benchmarks in PKB, p50 latency frequently falls below 1ms, it's not acceptable to report those numbers in 1ms granularity, e.g., local SSD based workloads, Aerospike workloads, etc. We need to change the measurement class configuration for those benchmark to use the RAW measurement, and write a PKB parser to consume the results (which is pretty simple).

First step is to create a complete list of benchmarks that need to move to use RAW class, then write a parser for them, then move over.