DF-0595: Michael MIC memcmp timing-channel demonstration
  iterations/case:   5000000
=== Test 1: principle (longer 256-byte buffers, SIMD-defeating) ===
  memcmp           first-byte-diff:    7.20 ns/call   last-byte-diff:   18.00 ns/call   ratio(last/first) = 2.501
  timingsafe_bcmp  first-byte-diff:  192.59 ns/call   last-byte-diff:  194.39 ns/call   ratio(last/first) = 1.009
=== Test 2: the actual kernel case (8-byte MIC tag) ===
  memcmp           first-byte-diff:    3.60 ns/call   last-byte-diff:    7.21 ns/call   ratio(last/first) = 2.006
  timingsafe_bcmp  first-byte-diff:    5.40 ns/call   last-byte-diff:    7.18 ns/call   ratio(last/first) = 1.329
  - Test 1 (>1 ratio for memcmp, ~1.0 for ct_bcmp): proves this libc's
  - Test 2 (noisy): at the real MIC length the signal is small and