Cyclic component models appear to be surprisingly slow

[axch]

Evidence: When originally building the satellites.bdb that we are distributing, turning the only two cyclic columns back to numerical seemed to accelerate analysis nearly 2x (240 iterations in 360 seconds rather than the 600 seconds it took with he cyclics) (one model per process).

[riastradh-probcomp]

This is a crosscat issue. Someone^TM needs to figure out what the von Mises component model is doing and why it is slow.

[fsaad]

I scanned the commit which created von Mises this morning and did not find anything glaring (log bessel computation?) We will need to run a profile.