investigate huge floating point errors in (flaky) tests

[sfindeisen]

Depending on the architecture and/or platform, we're sometimes getting test errors like:

FAIL (31.18s)
      test/common/TestMnistFCNN.hs:132:
      expected: 0.8991
       but got: 0.7424999999999999
      Use -p '/2 artificial 5 4 3 2 1 5 4 3 2 2393 1.0/' to rerun this test only.

This difference is huuge.

Let's try to investigate this and possibly also other similar test cases, understand what is going on, then fix the test case and/or the implementation.

Related issue: #46

[sfindeisen]

More failures:

$ cabal test --enable-optimization shortTestForCI
[...]
FAIL (10.29s)
      test/common/TestMnistFCNN.hs:198:
      expected: 0.7756000000000001
       but got: 0.7882
      Use -p '/VV artificial 5 4 3 2 1 5 4 3 2 0 18 2393 1.0/' to rerun this test only.
[...]
FAIL (7.47s)
      test/common/TestMnistFCNN.hs:275:
      expected: 0.8085
       but got: 0.8114
      Use -p '/LL artificial 5 4 3 2 1 5 4 3 2 0 3 3 2393 1.0 0.5/' to rerun this test only.
[...]
FAIL (13.25s)
      test/common/TestMnistFCNN.hs:548:
      expected: 0.8246
       but got: 0.8202
      Use -p '/S artificial 5 4 3 2 1 5 4 3 2 6 2393 1.0 0.5/' to rerun this test only.
[...]