Rewrite sparse dmatrix using callbacks. (#7092)

- Reduce dependency on dmlc parsers and provide an interface for users to load data by themselves.
- Remove use of threaded iterator and IO queue.
- Remove `page_size`.
- Make sure the number of pages in memory is bounded.
- Make sure the cache can not be violated.
- Provide an interface for internal algorithms to process data asynchronously.

tests/cpp/tree/gpu_hist/test_gradient_based_sampler.cu

@@ -27,7 +27,7 @@ void VerifySampling(size_t page_size,
   }
   gpair.SetDevice(0);
 
-  BatchParam param{0, 256, page_size};
+  BatchParam param{0, 256};
   auto page = (*dmat->GetBatches<EllpackPage>(param).begin()).Impl();
   if (page_size != 0) {
     EXPECT_NE(page->n_rows, kRows);
@@ -82,7 +82,7 @@ TEST(GradientBasedSampler, NoSamplingExternalMemory) {
   auto gpair = GenerateRandomGradients(kRows);
   gpair.SetDevice(0);
 
-  BatchParam param{0, 256, kPageSize};
+  BatchParam param{0, 256};
   auto page = (*dmat->GetBatches<EllpackPage>(param).begin()).Impl();
   EXPECT_NE(page->n_rows, kRows);