Pass infomation about objective to tree methods.

[trivialfis]

This PR adds a new struct for obtaining some information about the objective. The struct is not used yet but is required for the future development of categorical split. In #7214 , the optimal partitioning is only available for binary classification and regression. Copying inlined comment here:

 * The task field is useful for tree split finding, also for some metrics like auc.
 * While const_hess is useful for algorithms like adaptive tree where one needs to update
 * the leaf value after building the tree.  Lastly, knowing whether hessian is constant
 * can allow some optimizations like skipping the quantile sketching.

• Define ObjInfo.
• Pass it to tree methods.

[trivialfis]

I will pass it to metric once #7364 is approved or rejected to avoid conflicts. See #7386 for its usage in metrics.