AOAI results nits: fix order and _result column

[MilesHolland]

Apparently OAI likes to return their results in backwards (or possibly random) order, so I've changed the OAI results parsing to sort the results before merging them with the other evaluation results.

Also there was a bug in the pass/fail _result column code that would always result in the same value for that entire column. Changed to properly be a per-row result.

[Copilot]

Pull Request Overview

This PR fixes two issues with the AOAI evaluation results: ensuring that the results are merged in the correct order, and correcting a bug where the pass/fail _result column was producing a single value for all rows.

• The results dictionary is now initialized with an explicit "index" key to track row order.
• The _result column for each grader is now built by appending values per row from the raw results.

Comments suppressed due to low confidence (2)

sdk/evaluation/azure-ai-evaluation/azure/ai/evaluation/_evaluate/_evaluate_aoai.py:243

• [nitpick] Using the key 'index' might be confusing since it is commonly associated with a DataFrame's index. Consider renaming it to 'datasource_item_id' for better clarity.

listed_results = {"index": []}

sdk/evaluation/azure-ai-evaluation/azure/ai/evaluation/_evaluate/_evaluate_aoai.py:257

• [nitpick] The use of a magic number (50) to limit the length of result_column_name could be unclear to future maintainers. Consider defining and using a named constant for this limit.

if len(result_column_name) < 50: