Eliminate duplication between quantile_ensemble_stand and quantile_ensemble_flex

[brookslogan]

quantile_ensemble_stand and quantile_ensemble_flex contain a lot of similar code; for maintenance and development, and sharing performance improvements, it may be better to reduce or eliminate duplication.

Using #8, quantile_ensemble_flex should be significantly faster than before if all taus are in the same group. It may have similar or better performance to quantile_ensemble_stand and seems like with noncross=FALSE it would produce the same output if the class&shape of alpha is ignored. However, this makes no difference for the user-facing function quantile_ensemble, which dispatches to quantile_ensemble_stand if there is only a single tau group.

• decide whether to make a breaking change and have quantile_ensemble return a matrix alpha even when there is only a single tau group
• implement quantile_ensemble_stand by calling quantile_ensemble_flex with noncross=FALSE and other appropriate arguments, and then converting alpha or not depending on the above decision --- OR, if alpha will always be a matrix, consider eliminating quantile_ensemble_stand and maybe quantile_ensemble_flex altogether, putting the common code directly into quantile_ensemble
• add characterization tests for the single-tau-group case and make sure the above changes do not break them
• benchmark old and new implementations to make sure this is matching or improving the speed of the current quantile_ensemble_stand