Improve TDP handling for RASR decoding

[christophmluscher]

It is true that for Viterbi alignment and decoding we generally use the non-normalized default tdp values, however, for full-sum training, we generally introduce normalized loop/forward values that might be even estimated from an alignment or defined based on aomse heuristics, e.g. average phoneme length. In addition to this class, we could also introduce an enum class for TdpType, out of default, heuristic, and alignment-based. For the two latter types one could also have jobs that estimate values from an alignment or from transcription. Daniel Mann has already such jobs.

Originally posted by @Marvin84 in #110 (comment)

[christophmluscher]

just so we do not forget :)

[Marvin84]

I think @DanEnergetics has jobs for estimation of tdp from both an alignment and from the transcription. Maybe we could integrate this into the tdp class?

[DanEnergetics]

A while back I found this job for estimating tdps from alignment. Maybe this can be used. I have not tested it though and perhaps it can use some updates.
I don't have an estimate-from-transcription procedure for tdps, only for the prior.