Particle filter are too slow

[charlesknipp]

Consider a basic bootstrap filter; in the current state of this repository, it is fully functional and supports a wide variety of user defined models over a cacophony of distributions.

However, the current implementation relies on distributions being multivariate. Thus each particle set is a vector of static vectors, and because of the nature of Distributions.jl it is impossible to generalize rand(x::Sampleable) to fill an array given a univariate distribution. This becomes problematic because the filter must evaluate a boolean at each step of the filter, which is the obvious remedy to support univariate distributions.

A better solution would be to avoid the use of static arrays in the case of univariate distributions. This would reduce the computational strain added by multivariate distributions as well as reduce the overall number of allocations generated at each step.