Move prediction cache to Learner. (#5220)

* Move prediction cache into Learner.

* Clean-ups

- Remove duplicated cache in Learner and GBM.
- Remove ad-hoc fix of invalid cache.
- Remove `PredictFromCache` in predictors.
- Remove prediction cache for linear altogether, as it's only moving the
  prediction into training process but doesn't provide any actual overall speed
  gain.
- The cache is now unique to Learner, which means the ownership is no longer
  shared by any other components.

* Changes

- Add version to prediction cache.
- Use weak ptr to check expired DMatrix.
- Pass shared pointer instead of raw pointer.

tests/cpp/predictor/test_predictor.cc

@@ -0,0 +1,33 @@
+/*!
+ * Copyright 2020 by Contributors
+ */
+
+#include <cstddef>
+#include <gtest/gtest.h>
+#include <xgboost/predictor.h>
+#include <xgboost/data.h>
+
+#include "../helpers.h"
+#include "xgboost/generic_parameters.h"
+
+namespace xgboost {
+TEST(Predictor, PredictionCache) {
+  size_t constexpr kRows = 16, kCols = 4;
+
+  PredictionContainer container;
+  DMatrix* m;
+  // Add a cache that is immediately expired.
+  auto add_cache = [&]() {
+    auto *pp_dmat = CreateDMatrix(kRows, kCols, 0);
+    auto p_dmat = *pp_dmat;
+    container.Cache(p_dmat, GenericParameter::kCpuId);
+    m = p_dmat.get();
+    delete pp_dmat;
+  };
+
+  add_cache();
+  ASSERT_EQ(container.Container().size(), 0);
+  add_cache();
+  EXPECT_ANY_THROW(container.Entry(m));
+}
+}  // namespace xgboost