There's pathfinding and pathfinding. This thread is about exact algorithms for rectangular grids, not approximate algorithms for a continuous space.
There's pathfinding and pathfinding, and then there's pathfinding. Social network analysis also requires route-planning for degree-of-separation computation, but since in those cases you just have a digraph, not a [continuous or discrete] metric space, all you get to use there is Dijkstra.
(Anyone who knows better, I'd be delighted to find out I'm wrong—for the sake of my own search runtimes.)
They're already using a heuristic, I gave them another one to try. It's easy to discretize random samples to a grid.
No, A* heuristics (including extreme ones like jump point search) do not prejudice optimality of the found shortest path, the same that would be found by Dijkstra's algorithm without heuristics.
They are only "heuristic" with respect to the subordinate purpose of finding the optimal path cheaply (ignoring as many graph nodes as possible), while the RRT heuristic produces a suboptimal result (not the absolute shortest path, but the shortest path that goes through arbitrarily chosen nodes)