Fix PointEvaluator Differentiability

[Da-Be-Ru]

I require the use case to differentiate a point evaluation for H1P1 elements with respect to transformed point coordinates themselves, i.e. given some transformation $g: \mathbb{R}^d \rightarrow \mathbb{R}^d$ of the coordinate $\mathtt{x}$, I may have something like

$$f(\mathtt{x::Vector}) = \mathtt{PE.evaluate!}(g(\mathtt{x})) = \sum_k \alpha_k \cdot \varphi_k(g(\mathtt{x}))$$

and I need the gradient

$$D_xf(x) = \sum_k \alpha_k D_x\varphi_k(g(\mathtt{x}))D_x g(\mathtt{x}) = \sum_k \alpha_k \nabla\varphi_k(g(\mathtt{x}))\left[\partial_{x_j} g_l(\mathtt{x})\right]_{j,l}.$$

This required me to propagate the coordinate and coefficient types for the FEEvaluator which we missed to include when we separated those in the adjustments to the PointEvaluator to enable propagation of derivatives: #53.

I also added a test case for this on a reference triangle for different coefficient combinations $\mathbf{\alpha}$.

[chmerdon]

Ok, nice. Please add a bugfix version bump and a changelog entry.