feat: Support cross-database evaluation with SQLite ground truth

[arieljassan]

Overview

This PR adds support for running the BIRD benchmark (and similar cross-database SQL evaluations) on BigQuery while using local SQLite databases for ground truth reference execution.
When evaluating AI-generated queries on BigQuery against reference queries written in SQLite syntax (e.g., using STRFTIME or SQLite math functions), BigQuery cannot execute the reference queries directly. This update introduces a hybrid bridging mechanism that dynamically resolves reference answers from local SQLite files while evaluating generated queries on BigQuery.

Key Changes

• evalbench/databases/bigquery.py: Implements the required ensure_database_exists abstract method on BQDB to fulfill the base DB class contract.
• evalbench/scorers/sqlite_bridge.py & llmrater.py: Adds conditional ground truth resolution. When golden_error occurs and a hybrid judge is configured, LLMRater automatically fetches the true reference execution rows from the local SQLite database.
• hybrid_xa_judge.py: Adds a self-contained Execution Accuracy (XA) evaluator script for PythonScorer. It normalizes across engine data types (Decimal/Int64 vs float), ignores column header differences, and compares rows order-independently.

Verification

• Performed functional end-to-end benchmark tests across dual BIRD evaluation datasets (california_schools and card_games).
• Confirmed functional ground truth resolution from local SQLite reference tables during live BigQuery evaluation runs.
• Verified clean telemetry data warehousing into BigQuery (FLOAT64 score compatibility).

[arieljassan]

/gcbrun