* Support contextual-biasing for streaming model * The whole pipeline runs normally * Fix comments
* Don't reset model state and feature extractor on endpointing * support passing decoding_method from commandline * Add modified_beam_search to Python API * fix C API example * Fix style issues