Add LODR support to online and offline recognizers (#2026)

This PR integrates LODR (Level-Ordered Deterministic Rescoring) support from Icefall into both online and offline recognizers, enabling LODR for LM shallow fusion and LM rescore.

- Extended OnlineLMConfig and OfflineLMConfig to include lodr_fst, lodr_scale, and lodr_backoff_id.
- Implemented LodrFst and LodrStateCost classes and wired them into RNN LM scoring in both online and offline code paths.
- Updated Python bindings, CLI entry points, examples, and CI test scripts to accept and exercise the new LODR options.

sherpa-onnx/csrc/offline-lm-config.h

@@ -19,14 +19,23 @@ struct OfflineLMConfig {
   int32_t lm_num_threads = 1;
   std::string lm_provider = "cpu";
 
+  // LODR
+  std::string lodr_fst;
+  float lodr_scale = 0.01;
+  int32_t lodr_backoff_id = -1;  // -1 means not set
+
   OfflineLMConfig() = default;
 
   OfflineLMConfig(const std::string &model, float scale, int32_t lm_num_threads,
-                  const std::string &lm_provider)
+                  const std::string &lm_provider, const std::string &lodr_fst,
+                  float lodr_scale, int32_t lodr_backoff_id)
       : model(model),
         scale(scale),
         lm_num_threads(lm_num_threads),
-        lm_provider(lm_provider) {}
+        lm_provider(lm_provider),
+        lodr_fst(lodr_fst),
+        lodr_scale(lodr_scale),
+        lodr_backoff_id(lodr_backoff_id) {}
 
   void Register(ParseOptions *po);
   bool Validate() const;