New test “issue142” in 2.2.4 fails on aarch64, ppc64le, s390x

[musicinmybrain]

The output is identical on any of these three architectures:

# issue142
not ok 58 7 triangles when expected 4
ok 59 earcut deviation 0.0000000000000812% is less than 12.7549680371464884%
ok 60 libtess2 deviation 853.0670124916314307% is less than 853.0670130000000881%

See also #97; but rather than a small difference in deviation, this test is actually coming up with a different number of triangles.

Based on https://github.com/mapbox/earcut.hpp/issues/97#issuecomment-1032813710, I tried adding -ffp-contract=off to the compiler flags and found that doing so fixed this issue (in addition to #97).

[mourner]

cc @kkaefer