Update gemini response transform: calculate completion_tokens as the sum of candidatesTokenCount and thoughtsTokenCount to match OpenAI API expectations

[eulervoid]

Description:
With the OpenAI API, completion_tokens is meant to be the total number of completions tokens, including the ones further specified in completion_tokens_details, like reasoning_tokens. The current behavior of the google provider in portkey is to return them separately, which leads to incorrect usage calculations down the line.

This PR tries to fix this by:

• Calculating completion_tokens as the sum of candidatesTokenCount and thoughtsTokenCount to match the convention of the OpenAI API. I think this should probably also apply to audio_tokens, but i'm not as confident there, which is why i'm focusing on reasoning first with this PR.
• Adding this update for both google and google-vertex-ai providers.

Tests Run/Test cases added:

• No additional tests added
• Unfortunately, i wasn't able to run all the tests due to missing credentials

Type of Change:

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Refactoring (no functional changes)

[narengogi]

valid @eulervoid reviewing

[narengogi]

with vertex, I just want to verify if the behaviour is same as gemini for anthropic models on vertex also

[narengogi]

Looks good to me, let me update this PR with the relevant changes for anthropic on vertex as well

[VisargD]

@narengogi - Should we wait for your updates?

[narengogi]

fixing