Improve circumcenter computation performance

PR #202 closed Issue #201 demonstrated a speed-up in an end-to-end multigrid workflow simply by improving the performance of the circumcenter operation. That was due to type inference allowed by using static array type information. The circumcenter computation uses a generic inversion of a Cayley-Manger matrix for computing the circumcenter of a set of points.

This is nice since it generalizes over arbitrary dimension. However, there is potential for optimized versions of this operation, especially for the triangulated and tetrahedral mesh cases that we are interested in.

So, we should implement an optimized version(s) of the circumcenter operation that is more performant.