Porting issues from HMMBase.jl

[gdalle]

Gonna try to answer the dormant issues on HMMBase.jl, now that its successor HiddenMarkovModels.jl v0.4 can do pretty much everything I wanted it to do:

• Adding support for time-varying HMMs maxmouchet/HMMBase.jl#33 (@dmetivie, @kaandocal and @organic-chemistry): time-dependent HMMs are supported, see https://gdalle.github.io/HiddenMarkovModels.jl/stable/examples/temporal/
• Em Algorithm Speedup maxmouchet/HMMBase.jl#9 (@paschermayr): this version of EM is completely type-stable and nearly allocation-free, at least in the critical parts
• Benchmarks vs MS_HMMBase & question about messages_backwards_log maxmouchet/HMMBase.jl#11 and Multiple sequence with different length maxmouchet/HMMBase.jl#16 (@mmattocks, @sosuts and @cossio): multiple sequences are suppported, see https://gdalle.github.io/HiddenMarkovModels.jl/stable/examples/basics/#Multiple-sequences
• Example for multivariate features GMMHMM, include in docs maxmouchet/HMMBase.jl#40 (@rbeeli): mixture models can be supported if you have an algorithm for fitting, I'd be happy to discuss it