feat: add getResponse() method for accessing full response with usage data

[mattapperson]

Summary

• Add getResponse() method to ResponseWrapper for accessing complete response including usage data (inputTokens, outputTokens, cachedTokens, reasoningTokens, etc.)
• Update callModel documentation to include the new method
• Add comprehensive tests for response shape validation including all usage fields
• Add tests for provider parameter support

Usage

const response = client.callModel({
  model: "openai/gpt-4",
  input: "Hello!"
});

const fullResponse = await response.getResponse();
console.log(fullResponse.usage?.inputTokensDetails?.cachedTokens);

Test plan

• All 70 callModel e2e tests pass
• TypeScript compiles without errors
• New getResponse() tests verify usage shape including cachedTokens
• New provider parameter tests verify routing preferences work