Investigate the possibility of StaticArrays tableau methods

[ChrisRackauckas]

When I measured before, tableau-based methods were really really bad compared to hand-unrolled methods. But this was with Arrays and FixedSizeArrays. But now there's StaticArrays.jl, could it work?

The issue is that there's tons of specialization for a lot of the methods. Each of them "skip true zeros" in the tableaus (which, by the assumptions used to do the derivation, can be a lot in higher order methods). They have different error estimation schemes (DP8 and BS5 are good examples there). They have wildly different dense output schemes. Etc. So I don't know what we'll really get out of it because we'll still have to specialize a lot, but maybe it could be useful.

[ChrisRackauckas]

SDIRKs and Rosenbrocks can probably use it too. The issue is that doing this will make it so that .= doesn't put as much in a single call, which will hurt immensely hurt GPU kernels.