Fix (example/llm): workaround for new OPT default attention

src/brevitas_examples/llm/main.py

@@ -185,6 +185,8 @@ def main(args):
     kwargs = {"torch_dtype": dtype}
     if quant_sdpa_fx:
         kwargs["attn_implementation"] = "sdpa"
+    elif args.replace_mha:
+        kwargs["attn_implementation"] = "eager"
 
     if args.export_target == 'torch_qcdq':
         kwargs['torchscript'] = True