Update hf2mcore_deepseek_v3_moe.py

[lmc8133]

When setting expert-tensor-parallel-size=1, the weights of routed-experts will be saved repeatly during different tp_rank. It will consume ~1.3T*${TP} disk space, it's too huge!!

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}