Make randomly generated input consistent between different JDKs

[Fithos]

In some benchmarks, the input set is randomly generated before the first iteration using a fixed seed. While this ensures that, on the same JDK, the input set remains fixed, this assumption may not hold on different JDKs. I.e., the input set generated on JDK A may be different from the one generated on JDK B. Hence, the two workloads may differ. This is an important aspect when studying the determinism of the workloads. A good benchmark suite should minimize platform-specific behavior.

Reading the input set from a file or using a renaissance-specific RNG would remove this issue.

Benchmarks currently affected by this issue:

• als
• chi-square
• gauss-mix
• fj-kmeans
• scala-kmeans
• [please add more if you find some other workloads]