Implement robust regularization in 'survival:aft' objective (#5473)

* Robust regularization of AFT gradient and hessian

* Fix AFT doc; expose it to tutorial TOC

* Apply robust regularization to uncensored case too

* Revise unit test slightly

* Fix lint

* Update test_survival.py

* Use GradientPairPrecise

* Remove unused variables

doc/tutorials/aft_survival_analysis.rst

@@ -68,7 +68,7 @@ Note that this model is a generalized form of a linear regression model :math:`Y
 
   \ln{Y} = \mathcal{T}(\mathbf{x}) + \sigma Z
 
-where :math:`\mathcal{T}(\mathbf{x})` represents the output from a decision tree ensemble, given input :math:`\mathbf{x}`. Since :math:`Z` is a random variable, we have a likelihood defined for the expression :math:`\ln{Y} = \mathcal{T}(\mathbf{x}) + \sigma Z`. So the goal for XGBoost is to maximize the (log) likelihood by fitting a good tree ensemble :math:`\mathbf{x}`.
+where :math:`\mathcal{T}(\mathbf{x})` represents the output from a decision tree ensemble, given input :math:`\mathbf{x}`. Since :math:`Z` is a random variable, we have a likelihood defined for the expression :math:`\ln{Y} = \mathcal{T}(\mathbf{x}) + \sigma Z`. So the goal for XGBoost is to maximize the (log) likelihood by fitting a good tree ensemble :math:`\mathcal{T}(\mathbf{x})`.
 
 **********
 How to use

doc/tutorials/index.rst

@@ -18,6 +18,7 @@ See `Awesome XGBoost <https://github.com/dmlc/xgboost/tree/master/demo>`_ for mo
   monotonic
   rf
   feature_interaction_constraint
+  aft_survival_analysis
   input_format
   param_tuning
   external_memory