Upgrade data_path for query_with_files

[JingofXin]

Upgrade details are as follows:

1. Files are no longer transmitted with raw data encoding; instead, their paths are transmitted. The corresponding image or sound module is also responsible for decoding the data and saving it as a file.
2. Two API functions are exposed for encoding and decoding requests with files:
   • Exposed APIs:
     • encode_query_with_filepaths(query: str = None, files: List[str] = None) -> str
     • decode_query_with_filepaths(query_files: str) -> Union[dict, str]
   • Encoding format: lazyllm-query{‘query’: 'xxxxxx', 'files':['path1', 'path2']}
3. Enable bind to support merged encoding.
   lazyllm.TrainableModule('Mini-InternVL-Chat-2B-V1-5') | bind(_0, lazyllm_file='Describe this image')
4. Expose 1 API function for merging requests:
   lazyllm_merge_query(*args: str) -> str
5. A new passthrough method for file transfer is added: You can specify the module's ID, allowing only that module to receive the file.
   globals['lazyllm_files'][module._module_id] = ["path/to/file1"]
6. Add FileFormatter, and it will support encoding and decoding queries with files by setting the formatter to either decode or encode.

Assuming that the query and file have already been encoded, take them as input. Here are examples with sound and image, respectively: