[Refactor] Update the example usage for the @osmosis_rubric

[JoyboyBrian]

This pull request modernizes the rubric scoring workflow and improves usability, maintainability, and dataset support for the support conversation evaluation system. The most important changes include a refactor of the rubric scorer script to use a schema-driven config and dataset loader, updates to the workflow and documentation to support new usage patterns, and the introduction of a sample dataset for batch evaluation.

Rubric evaluation system refactor:

• Major refactor of reward_rubric.py to use schema-driven config loading, dataset records, and a simplified entrypoint (score_support_conversation). The script now loads YAML config and JSONL data, supports batch evaluation, and handles provider/model selection via config or environment.
• Updated reward_rubric_config.yaml to use a versioned schema with a rubrics[] array, separating rubric details and supporting multiple rubrics and default values.

Dataset and example improvements:

• Changed reward_rubric_example.json to use a flat structure (solution_str, original_input, ground_truth) instead of a message array, matching new dataset format.
• Added sample_data.jsonl as a JSONL dataset for batch rubric evaluation and CLI preview, with multiple conversation records.

Workflow and script updates:

• Updated GitHub Actions workflow (reward_rubric.yml) to call the new shell script (run_reward_rubric.sh) and trigger on changes to the example and scorer script, not just the config. [1] [2] [3]
• Simplified run_reward_rubric.sh to invoke the scorer as a Python module and accept CLI arguments for alternate data files.

Documentation enhancements:

• Expanded README.md to explain the new config schema, script usage, dataset format, and CLI options for previewing and evaluating rubrics. [1] [2] [3]

[JoyboyBrian]

The test fails is expected. Because we refactored the osmosis_ai sdk, which hasn't been merged and released yet.