[New Feature] fa3 支持flash mask (#3184)

* 支持flash mask

* 修改test_flash_mask

* 修改test.sh

custom_ops/setup_ops.py

@@ -497,6 +497,7 @@ elif paddle.is_compiled_with_cuda():
     if cc >= 90 and nvcc_version >= 12.0:
         # Hopper optmized mla
         sources += find_end_files("gpu_ops/mla_attn", ".cu")
+        sources += ["gpu_ops/flash_mask_attn/flash_mask_attn.cu"]
 
     setup(
         name="fastdeploy_ops",