BSR: Convert ARMA factors to polynomials along the Markov Chain

[Víctor de Buen Remiro]

Current ARMA simulator can only handle with AR and MA polynomials of degrees 1 or 2. So, polynomials with higher degrees must be factorized. If you want to simulate this structure

[[
  ARIMAStruct(1,1-0.01*B-0.01*B^2-0.01*B^3,1,1)
]]

you must to convert it in this one

[[
  ARIMAStruct(1,1-0.01*B-0.01*B^2, 1,1),
  ARIMAStruct(1,1-0.01*B,          1,1)
]]

Even, if you know that all roots must be real, then you can suplly this other one

[[
  ARIMAStruct(1,1-0.01*B, 1,1),
  ARIMAStruct(1,1-0.01*B, 1,1),
  ARIMAStruct(1,1-0.01*B, 1,1)
]]

If you want to simulate this structure

[[
  ARIMAStruct(1,1-0.01*B-0.01*B^2-0.01*B^3-0.01*B^4,1,1)
]]

you must to convert it in this one

[[
  ARIMAStruct(1,1-0.01*B-0.01*B^2, 1,1),
  ARIMAStruct(1,1-0.01*B-0.01*B^2, 1,1)
]]

The problem appairs when you try to make inference over the Markov chain of simulated parameters, due to there are no constraint to avoid that the simulations cross their identitities.

The solution is very simple: we only need to rebuilt the full polinomial of each simulation to get a Markov Chain of unmistakable coefficients.

[Víctor de Buen Remiro]

(In [1603]) Fixed #762

[Víctor de Buen Remiro]

Sorry, I've written "Fixed #762" instead of "Fixed #772" in last commit

[Víctor de Buen Remiro]

(In [2077]) Allows BSR to handle with ARMA factors with degree great than 2
Fixes #762