fix: Resolve unqualified columns in JOIN queries using schema info

[mingjerli]

Summary

• Fixes incorrect column lineage when unqualified columns are used in JOIN queries
• Uses sqlglot's qualify_columns optimizer with upstream table schemas to correctly resolve which table each column belongs to
• Fixes _extract_select_from_query to use dialect when serializing SQL, preventing function argument reordering (e.g., DATE_TRUNC)

Problem

When a SQL query joins multiple tables and columns are unqualified (no table prefix), the lineage builder would default to the first table, which was often incorrect. For example:

SELECT DATE_TRUNC(order_date, MONTH) as month
FROM analytics.user_metrics
JOIN staging.user_orders USING (user_id)

The order_date column would be incorrectly attributed to user_metrics instead of user_orders, causing the lineage edge staging.user_orders.order_date -> reports.monthly_revenue.month to be dropped.

Solution

1. Added _convert_to_nested_schema() helper to convert flat schema format to sqlglot's nested format
2. Added _qualify_sql_with_schema() helper to qualify unqualified columns using schema info before parsing
3. Modified RecursiveLineageBuilder.__init__ to qualify SQL before building lineage
4. Fixed _extract_select_from_query() to use dialect parameter when serializing SQL

Test plan

• Added 16 new tests in test_unqualified_column_resolution.py covering:
  • Schema conversion helpers
  • SQL qualification helpers
  • Single-query lineage with schema
  • Multi-query pipeline lineage (3-layer example)
  • Edge cases (columns in both tables, aggregates, fallback)
• All 778 existing tests pass (no regressions)
• Pre-commit checks pass

🤖 Generated with Claude Code