Speed up GEE mode

[jr-leary7]

• what it says in the title
• couple options for speedups ...
  • benchmark usage of Eigen matrix multiplication via RcppEigen versus doing so using the efficient Matrix package, and change to Matrix if it's better
  • replace usage of summary.gamlss() in backward_sel_WIC() with a getter function for the wald statistics based on the variance-covariance matrix and coefficient values, calculated like so: $\mathbf{W} = \symbf{\beta} / \sqrt{\text{Diag}(\mathbf{V})}$
  • see if any improvements can be had in score_fun_gee(), which is taking the most time currently (based on profiling with profvis)
  • replace usage of stats::lm.fit() in several scoring functions with RcppEigen::fastLmPure() as it seems to be about twice as fast (mean of 66mcs vs. 131mcs using microbenchmark with 500 trials)

[jr-leary7]

commit cff92ec addresses this to some extent ... still more to be done though

[jr-leary7]

summary: Eigen matrix multiplication is faster than that of Matrix, replaced usage of summary.gamlss() with more efficient version complete with error handling, and replaced stats::lm.fit() with RcppEigen::fastLmPure()