Compute metrics for all data once

Currently the correlations for All data are computed multiple times: first for all data in run_training_analyses and then during the analysis for each subgroup under compute_correlations_by_group.

For various reasons it would be helpful to keep the all_data on top of the subgroup analyses, but we should remove the computation from *_by_group (https://github.com/EducationalTestingService/rsmtool/blob/07f91264795dc8fb5de9518cb65ff4e1685836ef/rsmtool/analyzer.py#L833) and simply concatenate the output of _by_group with all_data computed previously.