CI Lab loose ends


 * Change taggp to properly handle untagging-based recursion elimination, both with real untagging and with default replacement.
 * Rewrite taggp so that it no longer eats stack frames and dies for deep recursion.
 * Conduct large scale tests of PushGP with tags on parity problems.
 * Develop dynamic problems (e.g. dynamic symbolic regression) and see if there's a relation between capacity for modularity (e.g. existence of tags) and problem solving performance on these problems.
 * Optimize Clojush. Using maps rather than multi-component-symbols for tag-related instructions may be one way to get significant speedups.
 * Upgrade Clojush to use Clojure 1.3.
 * Small Clojush bug (?): subst may be take its arguments the wrong way around, which may have implication for contains-subtree.
 * In tree GP, integrate tag-related functionality without adding depth to trees, via "intangibles" (kyles term) that are added in the same level as the function calls and "get out of the way" of evaluation. Or use Tom's suggestion of replacing functions with sets of functionality that can include both ordinary problem-related functions and tag-related functions.
 * Figure out if/when/how 100 runs may not be enough.