Enhancement Request: Allow Adding New Elements when Fine-Tuning with --init-model

### Background

It is a common approach to fine-tune a pre-trained AI model to related domain. e.g. transfer a QH9 pre-trained DPTB model to Li binded electrolyte molecules.


When utilizing the --init-model setting in DPTB for fine-tuning, the current implementation does not support the introduction of new elements (e.g., Lithium - Li in the mentioned case). This limitation restricts the model's adaptability to novel chemical domains.

### Describe the solution you'd like

add flag to allow the change of elements [here](https://github.com/deepmodeling/DeePTB/blob/57261bb10e5df0b26b9e0170c1becf31e954cc9f/dptb/entrypoints/train.py#L123)

### Additional Context

_No response_

Start at dptb/entrypoints/train.py around line 123 and trace how --init-model handles the model's element set. Compare the requested new-element fine-tuning path, including Li, with the existing training configuration. Done means a defined flag permits adding elements during initialization and the relevant training workflow is verified.

Written by the indexing model from the issue text.