Implement sin and cos

[tmcgrath325]

This implements sin and cos for GVar. Since the resulting distributions are not Gaussians, these will return a new GVar that approximates the target distribution by matching its mean and variance.

One thing to note is that the center of sin(a) is not necessarily equal to sin(a.center) (and the same goes for cos). A derivation and some tests of the result against empirical solutions can be found here.

[timholy]

Thanks! This looks great, the only things this needs are tests. I've added a general-purpose harness in #5 and improved it in #6. (Note that the implementation of exp also changed, and that might have an impact on some of the things you're using this for? In general I suspect the bounds are wider than previously.)

If you rebase this PR on current main, hopefully that should make the tests very easy to implement.