[Feature Request] Support for machine learning force field in OpenMM DMFF plugin

### Summary

Provide support for machine learning (ML) force field in OpenMM DMFF plugin. Present version does not yet support well, such as EANN and SGNN models.

### Motivation

The OpenMM DMFF plugin uses _save_dmff2tf.py_ script to transform a JAX model trained in DMFF to a tensorflow model used for OpenMM simulations. In this module, both classical and ADMP force field are considered. While the usage of ML force field in ther transformation is not clear. If ML force field are defined in XML file (referenced as [https://github.com/deepmodeling/DMFF/blob/master/docs/user_guide/4.4MLForce.md](url)), there will occur errors in running _save_dmff2tf.py_ script (detailed error message can be browsed in _Further Information_ part). The used ML models are EANN and SGNN.

### Suggested Solutions

1. Revise the potential generation function when using ML, and consider the situation of using ADMP and ML simultaneously.
2. Provide a specific instruction for the usage of ML forces in OpenMM DMFF Plugin docs.

### Further Information, Files, and Links

[reportbug.zip](https://github.com/deepmodeling/DMFF/files/13513636/reportbug.zip)
[test.zip](https://github.com/deepmodeling/DMFF/files/13513637/test.zip)
Above files record two different errors when using ML in this plugin.