Add ks-test (or some other likelihood) based DA

[Gabriel-p]

This DA attempts to find those stars in the cluster region that can generate the distribution more similar to the surrounding field regions, after N_memb stars were removed. The removed stars are then the best candidate cluster members.

The core of the DA is as follows:

cl_reg, fl_regs <-- Define a cluster region and `N_r` field regions
for fr in fl_regs:
    # Use some optimization algorithm
    while some_optimization_condition:
        # Optimize the selection of cl_sel so that p_value is maximized. When this happens,
        # we can not reject H0 which means both samples come from the same population.
        cl_sel <-- select N_memb stars from cl_reg
        cl_fr <-- remove cl_sel from cl_reg to create a 'clean' field region
        p_value = compare (cl_fr, fr) using the ks_test (or some other ML method)
    cl_sel_best <-- store the cl_sel that maximized the p_value
    p_value_best <-- assign the maximum p_value found to this selection
cl_reg_mp <-- average all the p_value_best for stars in all the cl_sel_best
              assigning 0. to stars not selected

Possible optimization method: https://stats.stackexchange.com/questions/27192/determine-a-subset-of-random-variables-that-are-most-correlated-with-the-whole?rq=1

SO question: https://stackoverflow.com/q/50879749/1391441

Explore greedy optimization methods and Greedy randomized adaptive search procedure.