Parameterize abstract filters

[jamgochiana]

https://github.com/sisl/GaussianFilters.jl/blob/1aa2a648021ea9c558b114f5d85065761a7d1625/src/ekf.jl#L9-L15

ExtendedKalmanFilter should be parameterized by DynamicsModel and ObservationModel types.

[zsunberg]

Parameterizing by the DynamicsModel and ObservationModel types will indeed yield significant performance improvement, but these parameters need not be present in function signatures, for example:

 function predict(b0::GaussianBelief, u::AbstractVector, filter::ExtendedKalmanFilter)

will have identical performance to

 function predict(b0::GaussianBelief, u::Vector{a}, filter::ExtendedKalmanFilter) where a<:Number

as you have now