Fixed single-GPU tests. (#4053)

- ./testxgboost (without filters) failed if run on a multi-GPU machine because
  the memory was allocated on the current device, but device 0
  was always passed into LaunchN

tests/cpp/common/test_gpu_compressed_iterator.cu

@@ -32,6 +32,7 @@ struct ReadSymbolFunction {
 };
 
 TEST(CompressedIterator, TestGPU) {
+  dh::safe_cuda(cudaSetDevice(0));
   std::vector<int> test_cases = {1, 3, 426, 21, 64, 256, 100000, INT32_MAX};
   int num_elements = 1000;
   int repetitions = 1000;