remove 2 bytes offset from align

[fmina]

@HannesSommer This fixes #176, but not sure if it generalizes to all platforms.

[ethzasl-jenkins]

Can one of the admins verify this patch?

[HannesSommer]

add to whitelist

[HannesSommer]

This is a pretty tough one. I don't believe your solution is right. It should lead to memory corruption in case the allocated isn't already aligned, because without the +16 or similar you might end up accessing memory in front of the allocated area (storage.bytes). The & makes the address always smaller. I wonder what lead to this problem on you side.
Could you easily add this as a unit test to the library? Then we can more easily test on different platforms. (jenkins is testing on ubuntu inidigo, xenial and some OSX )

Just add another member function below here that tests what you did for some fixed size eigen matrix:
https://github.com/ethz-asl/Schweizer-Messer/blob/master/numpy_eigen/test/numpy_eigen_tests.py#L111

The test_eigen function is currently skipped on Mac OS because something unknown is broken since ages. Maybe exactly what you've just discovered. I never had both time and motivation to look deeper into it. In general I'm afraid SM should officially drop Mac OS support unless somebody is volunteering to maintain it. I don't even have the means to to it.