Support hotword boosting feature and lexicon based decoding

[Sushmitha-Deva]

This pull request includes modifications that enable hotword boosting, lexicon-based decoding, and the build configuration for the ctcdecode library.

Hotword boosting:

• This feature is supported for both character and wpe based (allowed characters a-z and an apostrophe) ASR labels.
• Scoring logic for hotwords is inspired from the pyctcdecode package, where partial weight will be added to the path score containing the hotword tokens, and in case a complete hotword is not formed, it will reset the score to the original.
• Tests for using this feature is added in this PR

Lexicon based decoding:

• Lexicon based decoding ensures to penalize the path that's going to form an invalid word, thereby giving priority to paths containing valid spellings
• During decoding, it compares the beam path with the lexicon FST and then applies penality for unknown path formation.
• A constant negative value unk_score will be used as a penalty.
• Results indicate a 90% reduction in spelling mistakes after applying this decoding. Increase in unk_score, will lead to a 100% reduction.
• Create the lexicon FST using the build_fst tool . See tools/README.md for usage