Add --use-flash-attention flag. (#7223)

* Add --use-flash-attention flag.
This is useful on AMD systems, as FA builds are still 10% faster than Pytorch cross-attention.

comfy/model_management.py

@@ -930,6 +930,9 @@ def cast_to_device(tensor, device, dtype, copy=False):
 def sage_attention_enabled():
     return args.use_sage_attention
 
+def flash_attention_enabled():
+    return args.use_flash_attention
+
 def xformers_enabled():
     global directml_enabled
     global cpu_state