Send `GenerateContentRequest` in `CountTokensRequest`

[andrewheard]

• Updated the countTokens implementation to populate generateContentRequest instead of contents in the request body.
  • This provides a more accurate token count since, e.g., function declarations and system instructions from the request are now counted.
• Updated the GenerateContentRequest encoding implementation to serialize the model field.
  • Currently only needed for countTokens but is always serialized (it is ignored by the backend in generateContent and generateContentStream).