Add more convenience math functions and all tests pass

[kloudkl]

This PR fixed the build error in #159 and removed the code that might conflict with #161.

Two lessons learned:

1. Make sure to test again every time after rebasing.
2. Keep the codes in the branch that depends on it and nowhere else.

[shelhamer]

Thanks for the fixed PR! Checked that this builds and that tests pass.

Please run make lint and resolve the style complaints in math_functions.hpp, test_math_functions.cpp, and math_functions.cu.

[kloudkl]

All the lint errors except two about the CUDA kernel launch operator (linter wrongly takes <<< in macro definition as <<) have been fixed.

[shelhamer]

Add // NOLINT_NEXT_LINE(whitespace/operators) to the line above to grant an exception for the CUDA <<< and >>>. See https://github.com/BVLC/caffe/blob/dev/src/caffe/layers/pooling_layer.cu#L149 for an example.

[shelhamer]

I plan to merge this soon after #165, which is first since it is an older effort and broadens compatibility. This PR will need to be rebased then (it is not a clean merge now, either).

[kloudkl]

Yes, I am dealing with the trouble of a bunch of merge conflicts with dev.

[kloudkl]

Rebased, cpplinted and tested. Ready for merging.

[shelhamer]

Thanks @kloudkl!