Return AdapterFusion attentions using output_adapter_fusion_attentions argument

[calpt]

Example:

model = AutoAdapterModel.from_pretrained("bert-base-uncased")
model.eval()

tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")

model.add_adapter("a")
model.add_adapter("b")
model.add_adapter_fusion(["a", "b"])

model.active_adapters = Fuse("a", "b")

input_data = tokenizer("this is great", return_tensors="pt")

output_1 = model(**input_data, output_adapter_fusion_attentions=True)

print(output_1.adapter_fusion_attentions)