support optional mime_type in dspy.File.from_file_id

[dimitry12]

Support optional mime_type in dspy.File.from_file_id.

The main effect of this is: LiteLLM will not attempt to download file from file_id to determine its format.

This is critical when using Google Gemini and its Files API, because prior to this change the flow for multi-modal context was:

• use Files API to upload files;
• use each file's .uri as file_id and instantiate dspy.File.from_file_id which is then converted/"adaptered" into LiteLLM-schema dict,
• but LiteLLM attempts to determine format by downloading the file from URL contained in file_id. This is not possible because Google Gemini Files API is write-only, and the whole generation fails.

With this change we can pass Files API URIs as file_id and we can now provide optional mime_type explicitly:

dspy.File.from_file_id(f.uri, mime_type=f.mime_type)

preventing LiteLLM from attempting to download the file.

This change addresses:

• feat(File): add File type for handling file data #9014 (comment)
• feat(File): add File type for handling file data #9014 (comment)