Fix PseudoHuberLoss::LoadConfig()

[hcho3]

This pull request fixes the serialization of PseudoHuberLoss.

Update. The bug also appears when using the JSON serializer.

Minimal reproducer:

import os
from tempfile import TemporaryDirectory

import xgboost as xgb
from sklearn.datasets import load_diabetes

X, y = load_diabetes(return_X_y=True)
dtrain = xgb.DMatrix(X, label=y)
param = {
    "max_depth": 8,
    "eta": 1,
    "verbosity": 0,
    "objective": "reg:pseudohubererror",
}
xgb_model = xgb.train(
    param,
    dtrain,
    num_boost_round=10,
    evals=[(dtrain, "train")],
)
with TemporaryDirectory() as tmpdir:
    model_name = "model.json"
    #  model_name = "model.model"
    model_path = os.path.join(tmpdir, model_name)
    xgb_model.save_model(model_path)

Error message

xgboost.core.XGBoostError: [13:04:11] xgboost/src/common/json.cc:166: Object of type Null can not be indexed by string.
Stack trace:
  [bt] (0) xgboost/lib/libxgboost.so(dmlc::LogMessageFatal::~LogMessageFatal()+0x59) [0x7f385dc32bbd]
  [bt] (1) xgboost/lib/libxgboost.so(xgboost::Value::operator[](std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&)+0xa3) [0x7f385dd420c1]
  [bt] (2) xgboost/lib/libxgboost.so(xgboost::Json::operator[](std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&) const+0x36) [0x7f385dc34736]
  [bt] (3) xgboost/lib/libxgboost.so(xgboost::metric::PseudoErrorLoss::LoadConfig(xgboost::Json const&)+0x62) [0x7f385dfbac9c]
  [bt] (4) xgboost/lib/libxgboost.so(xgboost::LearnerImpl::EvalOneIter(int, std::vector<std::shared_ptr<xgboost::DMatrix>, std::allocator<std::shared_ptr<xgboost::DMatrix> > > const&, std::vector<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::allocator<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > > const&)+0x29b) [0x7f385df6e299]
  [bt] (5) xgboost/lib/libxgboost.so(XGBoosterEvalOneIter+0x24c) [0x7f385dc0fe0e]

[trivialfis]

Apologies for the bug. let me fix it.

[trivialfis]

I opened a different PR: #9575 , which adds the configuration to objective and have some tests.

The configuration passing is only implemented in the latest XGBoost (2.0) for metrics like ranking and quantile. I missed the pseudo huber function.