What is the final model of xgboost?

I have three questions about XGBoost.

1. What is the final model of XGBoost? That is, when I want to make a prediction, is the final prediction the average of all trees?
2. In R, how to check the prediction for each tree?
3. In R, how to interpret booster being gbtree and objective being reg:linear? Does that mean tree based model is used, and for leaf it is a linear regression model (rather than average)? If yes, what features did each leaf use?

Thanks!