Travis: Add macOS to the build matrix

[rht]

This fixes #359.

[rht]

Note: the version shipped in xcode10 is 3.7.

[coveralls]

Coverage decreased (-0.03%) to 94.547% when pulling 57ba55b on rht:macOS into 6ba4ba0 on QuantEcon:master.

[rht]

The first commit setup works fine (by install hdf5 via homebrew instead of conda), however, the machine was too slow to run the timing test. using conda is much faster

[mmcky]

thanks @rht -- this is great. I will check through the diff tomorrow and merge.

[rht]

OK, I found the test_rotated_hyper_ellipsoid to be non-deterministic though, in that sometimes on darwin it is fulfilled/failed, in that I can't write it as a test-expect-fail.

[mmcky]

hey @rht we have had tests running on linux for quite some time. I'm not sure why you're getting non-deterministic results. Are you sure its not just a tolerance issue?

[rht]

Otherwise, I can't explain the error in the commit "TestTicTacToc: Skip test_loop and test_timer for darwin". There was an error of ~1e-5.

[rht]

So, the test flake disappeared in the dust just like that.

[mmcky]

thanks @rht for the PR! I will merge this and take a look at tic tac toc issue for darwin when I get a chance. It's tested on the linux side anyway.