Remove unused config parameter from V1 XGBoost training function

Bug identified by cursor bot: The _xgboost_train_fn_per_worker function
had an unused 'config' parameter that was never utilized. This parameter
was misleading because:

1. All XGBoost training parameters are in 'xgboost_train_kwargs' which
   is bound via functools.partial
2. V1 XGBoostTrainer doesn't use train_loop_config pattern
3. The config parameter served no purpose and made the code confusing

Fix: Removed the config parameter and updated docstring to clarify this
is an internal V1 function where all parameters are pre-bound via partial,
unlike the V2 pattern where user functions receive train_loop_config.

This makes the V1 trainer's implementation pattern clearer and removes
confusion about parameter usage.

Signed-off-by: soffer-anyscale <[email protected]>

Author: soffer-anyscale

python/ray/train/xgboost/xgboost_trainer.py

@@ -264,7 +264,6 @@ def _configure_external_memory_params(config: dict):
 
 
 def _xgboost_train_fn_per_worker(
-    config: dict,
     label_column: str,
     num_boost_round: int,
     dataset_keys: set,
@@ -281,17 +280,21 @@ def _xgboost_train_fn_per_worker(
     the configuration. It manages checkpointing, dataset iteration, and
     training progress tracking.
 
+    Note:
+        This is an internal function used by the V1 XGBoostTrainer. All parameters
+        are bound via functools.partial before being passed to the base trainer,
+        unlike the V2 pattern where a user-defined function receives train_loop_config.
+
     Args:
-        config: XGBoost training configuration parameters. Should include
-            tree_method, objective, and evaluation metrics.
         label_column: Name of the label column in the dataset. Must exist
             in all datasets.
         num_boost_round: Target number of boosting rounds for training.
             When resuming from checkpoint, trains for remaining rounds.
         dataset_keys: Set of dataset names available for training. Should
             include at least TRAIN_DATASET_KEY.
-        xgboost_train_kwargs: Additional XGBoost training arguments such as
-            callbacks, verbose settings, etc.
+        xgboost_train_kwargs: XGBoost training parameters dictionary containing
+            tree_method, objective, eval_metric, and other XGBoost parameters.
+            This is passed directly to xgb.train().
         use_external_memory: Whether to use external memory for DMatrix creation.
             Required for large datasets that don't fit in RAM.
         external_memory_cache_dir: Directory for caching external memory files.