Add with_output version AppendAttention (#3302)

* get use_output from fd_config

* add clear TODO description

* add mask_offset para to align with develop

* fix bug

* fix use_output logic

* fix sot bug

fastdeploy/model_executor/layers/attention/flash_attn_backend.py

@@ -378,7 +378,7 @@ class FlashAttentionBackend(AttentionBackend):
             self.speculate_max_draft_token_num + 1,
             self.causal,
             self.speculative_method is not None,
-        )[0]
+        )
 
         if metadata.max_len_tensor_cpu[1] > 0:
             merge_prefill_decode_output(