Syncing Model Evaluation fields from gen-ai Public API

[aguan-hue]

Summary

• Sync recent model-evaluation and evaluation-dataset fields/enums into the public OpenAPI spec to match the live GenAI API. No new endpoints — additive fields only.

Adds

• Epochs (n-pass) on create/get run request and response, plus epoch result summary types
• reasoning_effort on candidate inference config
• Dataset paradigm on create/list dataset schemas, plus dataset_paradigm and has_ground_truth list filters
• CANDIDATE_MODEL_SOURCE_AGENT
• METRIC_CATEGORY_CONVERSATIONAL
• avg@k / pass@k / cons@k fields on metric result summaries