[misc] Add FlashMLA as a new option of VLLM_ATTENTION_BACKEND env (#14267)

Author: yangsijia-serena

vllm/envs.py

@@ -321,6 +321,7 @@ def maybe_convert_int(value: Optional[str]) -> Optional[int]:
     # - "XFORMERS": use XFormers
     # - "ROCM_FLASH": use ROCmFlashAttention
     # - "FLASHINFER": use flashinfer
+    # - "FLASHMLA": use FlashMLA
     "VLLM_ATTENTION_BACKEND":
     lambda: os.getenv("VLLM_ATTENTION_BACKEND", None),