Generic Seq2Seq Decoder

[ZhouW321]

might still run auto clang formatting

[curufinwe]

Plan for review (discussed between @SimBe195 and @curufinwe ):
We review and merge in 6 steps. All of the code can be reviewed but we will try to merge different parts separately.

1. small stuff (Bliss + Flf + Lattice + Speech (except aligner) + Tensorflow + Search/AdvancedTreeSearch+ Tools + Test), in addition we add a basic module without much functionality (Makefile + Module class) in order for the Makefile changes in other places to make sense.
2. LabelScorer + other changes in Nn
3. Search/
4. S2STS/LabelTree
5. S2STS rest except aligner
6. aligner

For each step we create a separate PR and once changes are done and merged we apply them back into this PR.

Time estimates:

1. 1-2 weeks
2. 1-2 weeks
3. 1 week
4. 1 week
5. 2-3 weeks
6. 1 week

Total: 7-10 weeks or around 3 months

Next deadline: step 1 should be finished April 7th