Refactor EhrSqlAnnotator for improved error handling and query management

[chakravarthik27]

This pull request updates the EHR SQL annotation and scenario components to improve SQL evaluation consistency, clarify instructions for SQL generation, and update data sources. The most significant changes ensure that SQL queries using current time are handled deterministically, update dataset and database URLs, and refine the logic for processing input data.

SQL Evaluation Improvements:

• Modified the annotation logic in ehr_sql_annotator.py to replace any usage of current_time in SQL queries with the fixed timestamp '2105-12-31 23:59:00', ensuring deterministic evaluation results. Also updated the expected result types from List[str] to List[Tuple[Any, ...]] for more accurate result handling. [1] [2]

Prompt and Adapter Specification Updates:

• Revised the SQL generation instructions in medhelm_run_specs.py to require the use of the fixed timestamp for current time, enforce stricter output formatting, and clarify that only a single SQL statement should be returned.

Dataset and Database Source Updates:

• Updated URLs in ehr_sql_scenario.py to point to the latest dataset, schema, and database sources, ensuring the scenario uses current and accessible resources.

Data Processing Logic:

• Improved input processing in ehr_sql_scenario.py to skip entries where the query is "null", preventing invalid data from being included in the scenario.

Minor Metadata Correction:

• Fixed a formatting issue in the model_metadata.yaml file for the upstage/solar-pro-241126 model's display name.

References:

• EHRSQL

[blidiselalin]

Looks good