Fix classifier dropout in AlbertForMultipleChoice (#13087)

Classification head of AlbertForMultipleChoice uses `hidden_dropout_prob` instead of `classifier_dropout_prob`.  This
is not desirable as we cannot change classifer head dropout probability without changing the dropout probabilities of
the whole model.

src/transformers/models/albert/modeling_albert.py

@@ -1282,7 +1282,7 @@ class AlbertForMultipleChoice(AlbertPreTrainedModel):
         super().__init__(config)
 
         self.albert = AlbertModel(config)
-        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+        self.dropout = nn.Dropout(config.classifier_dropout_prob)
         self.classifier = nn.Linear(config.hidden_size, 1)
 
         self.init_weights()