Add new constructor for Speech::ModelCombination to get config for specific model index

[larissakl]

This is a preparation of my one-pass model-combination algorithm which requires not only multiple label scorers, but also an individual lexicon per label scorer (and optionally an individual AM config). In this PR, a new constructor for Speech::ModelCombination is added which additional takes a modelIndex and retrieves the lexicon, AM and label scorer defined in the config for this index, e.g. like this:

[lib-rasr.search-algorithm.label-scorer-1]
type = ...

[lib-rasr.search-algorithm.lexicon-1]
file = ....

[lib-rasr.acoustic-model-1]
state-tying.type = ...

If no individual lexicon or AM is defined, it falls back to the global lexicon or AM.

In Core::Configuration, two new helper functions are added. The first checks if at least one resource under a certain selection is defined and the second returns the parent config for a selection.

I'm happy about any suggestions or ideas how to improve this code.