Fix TraceMe context propagation for some inference requests by replacing kPathwaysExecutor and kTfrtExecutor with a new enumkGenericModelExecutor and add back some tracing

PiperOrigin-RevId: 846455577

Author: tensorflower-gardener

tsl/profiler/lib/context_types.cc

@@ -54,6 +54,8 @@ const char* GetContextTypeString(ContextType context_type) {
       return "jax_serving";
     case ContextType::kScOffload:
       return "sparsecore_offload";
+    case ContextType::kGenericModelExecutor:
+      return "generic_model_executor";
   }
 }
 

tsl/profiler/lib/context_types.h

@@ -25,7 +25,7 @@ enum class ContextType : int {
   kGeneric = 0,
   kLegacy,
   kTfExecutor,
-  kTfrtExecutor,
+  kTfrtExecutor [[deprecated]],
   kSharedBatchScheduler,
   kPjRt,
   kAdaptiveSharedBatchScheduler,
@@ -35,12 +35,13 @@ enum class ContextType : int {
   kBatcher,
   kTpuStream,
   kTpuLaunch,
-  kPathwaysExecutor,
+  kPathwaysExecutor [[deprecated]],
   kPjrtLibraryCall,
   kThreadpoolEvent,
   kJaxServingExecutor,
   kScOffload,
-  kLastContextType = ContextType::kScOffload,
+  kGenericModelExecutor,
+  kLastContextType = ContextType::kGenericModelExecutor,
 };
 
 // In XFlow we encode context type as flow category as 6 bits.