feat: support resuming existing W&B runs

[Zhuang-Zhuang-Liu]

Summary:

• add --wandb-resume argument for W&B run resumption
• pass wandb run id and resume mode into wandb.init when provided
• update the qwen3 4b RL script example to show how to resume the same W&B run

Example:
Uncomment the following in WANDB_ARGS when resuming:

• --wandb-run-id <existing_run_id>
• --wandb-resume must