docs(transfer-learning): clarify simplify vs transfer learning

[njzjz-bot]

Problem

• The tutorial text currently implies that transfer learning in DP-GEN is essentially the "simplify" module, which can confuse users (e.g., thinking simplify is required for transfer learning).

Change

• Reword the "How to implement transfer-learning" intro to clarify that simplify is one workflow that facilitates fine-tuning via init-model, and that transfer learning/fine-tuning can also be done directly in DeePMD-kit using an existing model (init_model) on a smaller target dataset.
• Minor typo fix: "unit" -> "once".

Authored by OpenClaw (model: gpt-5.2)