`predict.xgboost` silently ignores `predcontrib = T`

With `xgboost` 3.2.1.1:

```R
m = xgboost::xgboost(
x = mtcars[, c("cyl", "disp", "hp")],
y = mtcars$mpg)
str(predict(m,
mtcars[, c("cyl", "disp", "hp")],
predcontrib = T))
# Named num [1:32] 21 21 22.8 21.4 18.7 ...
# - attr(*, "names")= chr [1:32] "Mazda RX4" "Mazda RX4 Wag" "Datsun 710" "Hornet 4 Drive" ...
```

What the user should do is say `type = "contrib"`, because XGBoost's two `predict` methods have inconsistent interfaces for this. But it's hard to catch what's going on because `predict.xgboost` produces no warning or error for unrecognized parameters, and `predcontrib = T` used to work in this context in previous versions of XGBoost.

Start by examining the R `predict.xgboost` implementation and how it handles unrecognized prediction parameters, then compare its interface with the other `predict` method. Reproduce the `predcontrib = T` example and verify that the completed behavior either reports the unsupported argument clearly or preserves the previously supported usage.

Written by the indexing model from the issue text.