Drop default layerwise_checkpoint_dir from max PTQ recipe

Max calibration is fast enough that checkpointing each layer adds
unnecessary I/O and disk usage. Comment explains why it is omitted.

Signed-off-by: realAsma <akuriparambi@nvidia.com>

Author: realAsma

modelopt_recipes/general/ptq/nvfp4_experts_only-fp8_kv.yaml

@@ -19,8 +19,8 @@ metadata:
 quantize:
   algorithm:
     method: max
+    # Max calibration is fast and does not typically need checkpointing.
     layerwise: true
-    layerwise_checkpoint_dir: output/layerwise_ckpts/
   quant_cfg:
     - quantizer_name: '*'
       enable: false