fix(validate): case-insensitive aesthetic-clause column lookup

[dataders]

Summary

VISUALISE clauses in .gg.sql files currently fail validation against case-preserving warehouses (Snowflake, Databricks, BigQuery quoted IDs) when the user writes lowercase column references. The columns come back from the warehouse in their declared case (typically UPPERCASE), and ggsql's column-name resolver does exact-case comparison, producing:

Validation error: Column 'table_name' referenced in aesthetic 'pos1' does not exist.
Available columns: TABLE_CATALOG, TABLE_SCHEMA, TABLE_NAME, ROW_COUNT

DuckDB folds identifiers to lowercase by default, so the bug only bites case-preserving sources.

Fix

Case-insensitive column resolution with an exact-match-first preference:

1. execute::canonicalize_column_references runs after the source schema is fetched and rewrites layer mappings, remappings, and partition_by columns to the schema's canonical case. This is what actually unblocks the SQL — without canonicalization, the generated SQL emits \"table_name\" AS \"__ggsql_aes_pos1__\" which a case-preserving warehouse rejects because the real column is TABLE_NAME. The user's original casing is preserved on AestheticValue::Column.original_name so axis labels render the user's intent.

2. writer::vegalite::layer::validate_layer_columns also uses a tolerant case-insensitive lookup as defense in depth, in case any code path slips past canonicalization.

Resolution rules in both layers:

• Exact-case match wins outright (no behavior change for DuckDB / lowercase schemas).
• Otherwise, case-insensitive lookup; a single match is accepted, multiple matches raise a clear ambiguity error so we never silently guess.
• No match → the existing "column does not exist" error is preserved.

Evidence

This bug was independently reproduced from two different reader stacks against real Snowflake — confirming the root cause is ggsql-internal, not reader-side conversion:

• XdbcReader (fusion-stack via dbt-xdbc): hit the same Validation error: Column 'table_name' … against analytics_dev.information_schema.tables.
• AdbcReader (clean-room Apache ADBC, no fusion dependencies): hit identical symptom from the same query.

Both readers correctly preserve the warehouse's casing; the validator was the only mismatched component.

Test plan

• New unit test test_case_insensitive_column_lookup — DataFrame columns TABLE_NAME, ROW_COUNT; aesthetics reference table_name (lowercase) and ROW_COUNT (exact case) — write succeeds.
• New unit test test_case_insensitive_lookup_ambiguity — DataFrame has both Foo and FOO; aesthetic references foo — error includes "ambiguous".
• Existing test_missing_column_error still passes (no regression to genuinely-missing-column path).
• Full cargo test --lib green: 1365 passed, 0 failed, 1 ignored.
• Manual verification against a Snowflake-backed .gg.sql (left to maintainers — needs warehouse credentials).

🤖 Generated with Claude Code