fix: normalize_column_name does not account for output case settings on non-Snowflake DBs

[ota2000]

What

normalize_column_name() does not perform case conversion on non-Snowflake databases, which breaks idempotency in inject_missing_columns, remove_columns_not_in_database, and synchronize_data_types when combined with output-to-upper or output-to-lower settings.

Reproduction scenario

PostgreSQL + output-to-upper: true:

1. First run: DB returns zebra → get_columns() key is zebra (normalize is a no-op) → inject_missing_columns adds node.columns["ZEBRA"] (output-to-upper applied)
2. Second run: current_columns = {normalize("ZEBRA", "postgres")} = {"ZEBRA"}, incoming_name = "zebra" → "zebra" not in {"ZEBRA"} → True → re-added (overwritten)

Root cause

Both current_columns and incoming_columns are compared using normalize_column_name, but normalize_column_name is case-preserving on non-Snowflake databases. This means column keys transformed by output-to-upper/output-to-lower won't match the DB-derived normalized names.

Affected locations

• src/dbt_osmosis/core/transforms.py: inject_missing_columns (L333-336)
• src/dbt_osmosis/core/transforms.py: remove_columns_not_in_database (L379-382)
• src/dbt_osmosis/core/transforms.py: synchronize_data_types (L535-538)

Notes

• A naive case-insensitive comparison could break databases that distinguish quoted columns ("Foo" vs "foo")
• This likely requires either rethinking normalize_column_name or introducing a separate comparison normalizer that accounts for output case settings