make advection memory (largely) independent of number of species

[zingale]

We should be able to reformulate the CTU method to do the species updates "on demand" by upwinding based on the contact speed (for non-HLL solvers), and therefore not need to store the fluxes of the species. We'll use this issue to explore this idea.