docs: remove undocumented --make_moe_param_leaf_module flag (#3)

This flag is documented in finetune/README.md:146 and
finetune/README_CN.md:146 but not defined in
finetune/deepspeed_support/train.py (ModelArguments/TrainingArguments).
HfArgumentParser would reject it as unrecognized arguments.
Remove from docs until implemented. Also fix line-break typo in CN
where --gradient_checkpointing and --learning_rate were concatenated.