Track token scores (#571)

* add export of per-token scores (ys, lm, context)

- for best path of the modified-beam-search decoding of transducer

* refactoring JSON export of OnlineRecognitionResult, extending pybind11 API of OnlineRecognitionResult

* export per-token scores also for greedy-search (online-transducer)

- export un-scaled lm_probs (modified-beam search, online-transducer)
- polishing

* fill lm_probs/context_scores only if LM/ContextGraph is present (make Result smaller)

sherpa-onnx/csrc/online-recognizer-transducer-impl.h

@@ -69,6 +69,10 @@ static OnlineRecognizerResult Convert(const OnlineTransducerDecoderResult &src,
     r.timestamps.push_back(time);
   }
 
+  r.ys_probs = std::move(src.ys_probs);
+  r.lm_probs = std::move(src.lm_probs);
+  r.context_scores = std::move(src.context_scores);
+
   r.segment = segment;
   r.start_time = frames_since_start * frame_shift_ms / 1000.;