When we load data, we make sure that the number of benchmarks in the
features data is the same as the number of JMH benchmarks.

We also check that we didn't lose any benchmarks when merging the two
data-sets together (which could happen silently if you merge two files
about different benchmarks)