Add floating point hardware checking for Intel on GNU compilers...When using the -check function (the default) it is enabled for Kinship
computation and LM/LMM up to individual SNP computation. This means
there can no longer be NaN values for matrices that are reused for every
SNP, but it is possible to have NaN for individual SNPs.
Fixes #161
Travis-ci:...- Disabled gcc-6 since we develop with later tools anyway
- Turned the release test into a simple integration test
- Adding MacOSX on Travis-ci
Tests: Adding tests for Plink w. LOCO
Safety: Introduce strtok_safe to get rid of segfaults