feat: #1831 Add opt-in cost tracking for LiteLLM models

[habema]

Resolves #1831

Adds cost tracking for LiteLLM models when users enable it via ModelSettings(track_cost=True).

Changes

• Added cost field to Usage class
• Added track_cost setting to ModelSettings. Opt-in (defaults to False)
• LiteLLM cost extraction:
  • Non-streaming: reads from _hidden_params["response_cost"]
  • Streaming: calculates via litellm.completion_cost()
  • Includes cached & reasoning tokens for accurate pricing

[seratch]

Unlocking the access itself is fine, but we'd like to hold off relying on an internal modules within a different package.

[habema]

litellm.completion_cost() is not used for both streaming and non-streaming responses.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.