Subtype of hmm.B and automatic type for outputs

[dmetivie]

In the HMM structure, I think it is best practice to have
B::Vector{<:Distribution{F}} instead of B::Vector{Distribution{F}} see for example the responses to my question here.

So far the output type is always Float64 e.g. y = Vector{Float64}(undef, length(z)). I believe for discrete distribution this is not ideal.
Something like y = Vector{eltype(eltype(hmm.B))}(undef, length(z)) would detect from the distribution in B the type.
This looks pretty ugly though. Maybe the type of the output distribution could be included in the hmm ?

Moreover, I did not change that but all matrix (in algorithm) type are Float64, this could be changed to something related to the T used in HMM{F,T}?

Thanks for the package.