[Hygon] Support qwen3-0.6B training

[Leewzh]

This code is for qwen3-0.6B large-scale model training，below are some changes made：

1. modified /FlagScale/build/Hygon_BW1000/FlagScale/flagscale/train/train.py
2. modified /FlagScale/build/Hygon_BW1000/FlagScale/third_party/Megatron-LM/megatron/training/arguments.py
3. modified /FlagScale/build/Hygon_BW1000/FlagScale/examples/qwen3/conf/train.yaml
4. modified /FlagScale/build/Hygon_BW1000/FlagScale/examples/qwen3/conf/train/0_6b.yaml
5. add /FlagScale/build/Hygon_BW1000/FlagScale/env.sh

[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.

---

liwzh7 seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}