Adds the JNI call for Cuda.deviceSynchronize (#10839)

Running tests locally, but putting this up as WIP for now.

Discussing with @jlowe a solution to NVIDIA/spark-rapids#4818 could involve `cudaDeviceSynchronize.` I noticed that's not in our JNI exposed calls, so I am adding it here.

Authors:
  - Alessandro Bellina (https://github.com/abellina)

Approvers:
  - Jason Lowe (https://github.com/jlowe)

URL: #10839

java/src/main/java/ai/rapids/cudf/Cuda.java

@@ -596,4 +596,10 @@ public static void multiBufferCopyAsync(long [] destAddrs,
    * no effect.
    */
   public static native void profilerStop();
+
+  /**
+   * Synchronizes the whole device using cudaDeviceSynchronize.
+   * @note this is very expensive and should almost never be used
+   */
+  public static native void deviceSynchronize();
 }

java/src/main/native/src/CudaJni.cpp

@@ -390,4 +390,12 @@ JNIEXPORT void JNICALL Java_ai_rapids_cudf_Cuda_profilerStop(JNIEnv *env, jclass
   CATCH_STD(env, );
 }
 
+JNIEXPORT void JNICALL Java_ai_rapids_cudf_Cuda_deviceSynchronize(JNIEnv *env, jclass clazz) {
+  try {
+    cudf::jni::auto_set_device(env);
+    CUDF_CUDA_TRY(cudaDeviceSynchronize());
+  }
+  CATCH_STD(env, );
+}
+
 } // extern "C"