Refactor configuration [Part II]. (#4577)

* Refactor configuration [Part II].

* General changes:
** Remove `Init` methods to avoid ambiguity.
** Remove `Configure(std::map<>)` to avoid redundant copying and prepare for
   parameter validation. (`std::vector` is returned from `InitAllowUnknown`).
** Add name to tree updaters for easier debugging.

* Learner changes:
** Make `LearnerImpl` the only source of configuration.

    All configurations are stored and carried out by `LearnerImpl::Configure()`.

** Remove booster in C API.

    Originally kept for "compatibility reason", but did not state why.  So here
    we just remove it.

** Add a `metric_names_` field in `LearnerImpl`.
** Remove `LazyInit`.  Configuration will always be lazy.
** Run `Configure` before every iteration.

* Predictor changes:
** Allocate both cpu and gpu predictor.
** Remove cpu_predictor from gpu_predictor.

    `GBTree` is now used to dispatch the predictor.

** Remove some GPU Predictor tests.

* IO

No IO changes.  The binary model format stability is tested by comparing
hashing value of save models between two commits

tests/cpp/tree/test_gpu_exact.cu

@@ -16,7 +16,7 @@ TEST(GPUExact, Update) {
   std::vector<Arg> args{{"max_depth", "1"}};
 
   auto* p_gpuexact_maker = TreeUpdater::Create("grow_gpu", &lparam);
-  p_gpuexact_maker->Init(args);
+  p_gpuexact_maker->Configure(args);
 
   size_t constexpr kNRows = 4;
   size_t constexpr kNCols = 8;

tests/cpp/tree/test_gpu_hist.cu

@@ -384,10 +384,11 @@ void TestHistogramIndexImpl(int n_gpus) {
     {"max_leaves", "0"}
   };
 
-  LearnerTrainParam learner_param(CreateEmptyGenericParam(0, n_gpus));
-  hist_maker.Init(training_params, &learner_param);
+  GenericParameter generic_param(CreateEmptyGenericParam(0, n_gpus));
+  hist_maker.Configure(training_params, &generic_param);
+
   hist_maker.InitDataOnce(hist_maker_dmat.get());
-  hist_maker_ext.Init(training_params, &learner_param);
+  hist_maker_ext.Configure(training_params, &generic_param);
   hist_maker_ext.InitDataOnce(hist_maker_ext_dmat.get());
 
   ASSERT_EQ(hist_maker.shards_.size(), hist_maker_ext.shards_.size());

tests/cpp/tree/test_prune.cc

@@ -37,7 +37,7 @@ TEST(Updater, Prune) {
   std::vector<RegTree*> trees {&tree};
   // prepare pruner
   std::unique_ptr<TreeUpdater> pruner(TreeUpdater::Create("prune", &lparam));
-  pruner->Init(cfg);
+  pruner->Configure(cfg);
 
   // loss_chg < min_split_loss;
   tree.ExpandNode(0, 0, 0, true, 0.0f, 0.3f, 0.4f, 0.0f, 0.0f);

tests/cpp/tree/test_quantile_hist.cc

@@ -236,7 +236,7 @@ class QuantileHistMock : public QuantileHistMaker {
   explicit QuantileHistMock(
       const std::vector<std::pair<std::string, std::string> >& args) :
       cfg_{args} {
-    QuantileHistMaker::Init(args);
+    QuantileHistMaker::Configure(args);
     builder_.reset(
         new BuilderMock(
             param_,

tests/cpp/tree/test_refresh.cc

@@ -37,7 +37,7 @@ TEST(Updater, Refresh) {
   tree.Stat(cleft).base_weight = 1.2;
   tree.Stat(cright).base_weight = 1.3;
 
-  refresher->Init(cfg);
+  refresher->Configure(cfg);
   refresher->Update(&gpair, dmat->get(), trees);
 
   bst_float constexpr kEps = 1e-6;

tests/cpp/tree/test_split_evaluator.cc

@@ -32,7 +32,7 @@ TEST(SplitEvaluator, Interaction) {
     }
     std::vector<int32_t> solutions{4, 5};
     ASSERT_EQ(accepted_features.size(), solutions.size());
-    for (int32_t f = 0; f < accepted_features.size(); ++f) {
+    for (size_t f = 0; f < accepted_features.size(); ++f) {
       ASSERT_EQ(accepted_features[f], solutions[f]);
     }
   }