XGBClassifier params eval_metric don't support multiple custom metrics

eval_metric=f1_score is okay,
eval_metric='auc' is okay,
eval_metric=['rmse','auc'] is ok,
eval_metric=[f1_score,'auc'] don't work.
sklearn.metrics or custom metrics function can't use with bulit-in functions, nor can use multiple like eval_metric=[f1_score,roc_auc_score].
but lightgbm can do it.

Start at the XGBClassifier entry point and its eval_metric handling. Reproduce the cases described with sklearn.metrics or custom metric functions, including a custom function mixed with 'auc' and multiple custom metrics; done means these combinations work consistently alongside the existing single and built-in metric cases.

Written by the indexing model from the issue text.