Implement a much faster cbrt function.

[drwells]

I profiled the new Holzapfel-Ogden code and was surprised that the exponentials are much less expensive than computing cube roots. In fact, calling std::cbrt() takes up a third of the total stress function evaluation time.

This needs to be checked some more (replacing parts of the standard library is always risky) but swapping out std::cbrt() for our own custom windowed implementation essentially eliminates the cost of calling that function. Since we spend 50% of our time in the (old) heart model code calling these functions it is definitely worth investigating.