Use matrix for gradient. (#9508)

- Use the `linalg::Matrix` for storing gradients.
- New API for the custom objective.
- Custom objective for multi-class/multi-target is now required to return the correct shape.
- Custom objective for Python can accept arrays with any strides. (row-major, column-major)

R-package/R/utils.R

@@ -154,7 +154,14 @@ xgb.iter.update <- function(booster_handle, dtrain, iter, obj) {
     pred <- predict(booster_handle, dtrain, outputmargin = TRUE, training = TRUE,
                     ntreelimit = 0)
     gpair <- obj(pred, dtrain)
-    .Call(XGBoosterBoostOneIter_R, booster_handle, dtrain, gpair$grad, gpair$hess)
+    n_samples <- dim(dtrain)[1]
+    # We still require row-major in R as I'm not quite sure sure how to get the stride of
+    # the matrix in C.
+    gpair$grad <- matrix(gpair$grad, nrow = n_samples, byrow = TRUE)
+    gpair$hess <- matrix(gpair$hess, nrow = n_samples, byrow = TRUE)
+    .Call(
+      XGBoosterBoostOneIter_R, booster_handle, dtrain, iter, gpair$grad, gpair$hess
+    )
   }
   return(TRUE)
 }