Adding in option to run jackknife standard errors for gee_test (and testing this capability)

[svteichman]

This PR does the following:

• adds the argument use_jack_se (default is FALSE) to gee_test()
• adds a function jackknife_se() and helper function update_model() that compute jackknife standard errors, with an optional argument of id which calculates a clustered jackknife se estimator (this is appropriate for our gee test function)
• adds tests of these new functionalities

One thing to note is that this jackknife implementation is slow (when benchmarked against the jackknife standard errors from geepack::geeglm()), which makes sense because it relies on an sapply call instead of calling cpp code. I think this shouldn't be a problem because the use case is when there is large bias for sandwich se's for a small number of clusters (according to the comment in line 43-45 of "gee_test.R"). If this all looks good, next steps are probably to add in this option to enviromtx so that someone could decide when calling fit_mgx_model whether they want sandwich or clustered jackknife standard errors based on the number of clusters.