ProjectIncomp should do Sobolev projection

[Jacob Hinkle]

In our fluid kernels, if we have divergenceFree=True, we do a projection against a complex vector representing the divergence operator. The projection we use though is implicitly using the L2 inner product. Since we have the pointwise 3x3 matrix representing our metric, we should use that for the projection instead.

One point to be aware of is that currently we handle a special case for degenerate kernels, for instance so that we can evaluate the Laplacian and invert it subject to zero boundary conditions. We just need to be aware of this possibility and implement the true \dot{H} projection.