Display $chi^2$ for shown bins in two-dimensional plots

[epelaaez]

This PR addresses #481 by making it so each slice and projection in a set of two-dimensional plots shows the $\chi^2/\text{nbins}$ only for the displayed bins rather than for the whole fit. It also adds the full $\chi^2$ to the two-dimensional heat maps.

For each 2D slice or projection, a matrix $P$ maps the original flattened bins onto the bins displayed on that page as $s' = Ps$, where $s$ contains the original bins and $s'$ the ones displayed. Each row of $P$ is one displayed bin. It selects or sums over bins from $s$ to construct $s'$.

For PROchi and PROCNP, the residual and covariance are projected like $\delta' = P\delta, M' = PMP^T$ and $\chi^2 = \delta'^T(M')^{-1}\delta'$. For the Poisson metric, the prediction and data are projected first as $\mu' = P\mu, n' = Pn$, and the Poisson statistic is evaluated using the projected counts.

A few examples: