loadbearer v0.3.1

• Single-threaded subtests are pinned to one core while they run, so the OS
  scheduler can't bounce the measurement between core types (P/E cores on Intel
  hybrid CPUs, big.LITTLE on ARM) part-way through. On a 13th-gen Intel laptop
  this cut single-threaded memory-bandwidth run-to-run variance from ±40–60%
  to a few percent. Each such iteration runs on a throwaway thread pinned to the
  fastest core (highest rated frequency on Linux; the first reported core
  elsewhere). All-core subtests are unaffected. Adds the core_affinity crate.

Full Changelog: v0.3.0...v0.3.1