[Vertex AI] Add Developer API encoding CountTokensRequest

[andrewheard]

Add support for encoding CountTokensRequest in the format required by both the Vertex AI API and the Developer API.

• Add unit tests for encoding CountTokensRequest

#no-changelog

[gemini-code-assist]

Code Review

The pull request introduces support for encoding CountTokensRequest for both Vertex AI API and Developer API. The changes involve modifying the CountTokensRequest and GenerateContentRequest structures and adding a new integration test. Overall, the code seems well-structured and addresses the objective. However, there are a few areas that could be improved for clarity and maintainability.

Summary of Findings

• Encoding Logic in CountTokensRequest: The encoding logic in CountTokensRequest could be made more readable by extracting the encoding logic for each service type into separate functions.
• Redundant Properties in CountTokensRequest: The options and apiConfig properties in CountTokensRequest are now redundant as they are accessed through generateContentRequest. Consider removing them to simplify the struct.
• Missing Unit Tests: There are no unit tests for the encoding logic in CountTokensRequest. Adding unit tests would ensure that the encoding is correct for both Vertex AI and Developer APIs.

Merge Readiness

The pull request is almost ready for merging. However, I recommend addressing the feedback provided, particularly regarding the encoding logic in CountTokensRequest and adding unit tests. I am unable to directly approve this pull request, and recommend that others review and approve this code before merging. Addressing the high severity issues is crucial before merging.