feat: implicit VARCHAR-to-numeric cast for aggregate functions

[mattotodd]

Problem

Snowflake implicitly casts VARCHAR columns to numeric when used in aggregate functions like SUM(), AVG(), MEDIAN(), etc. DuckDB is strict and rejects these:

Binder Error: No function matches the given name and argument types 'sum(VARCHAR)'.

This is a common pattern when data is loaded via CSV (everything comes in as VARCHAR) and then aggregated without explicit casts.

Solution

Add a numeric_agg_implicit_cast transform that wraps arguments to numeric-only aggregate functions with TRY_CAST(... AS DOUBLE), matching Snowflake's implicit coercion behavior.

• Only applies to numeric-only aggregates: SUM, AVG, MEDIAN, VARIANCE, STDDEV, etc.
• Does NOT affect COUNT, MIN, MAX, ARRAY_AGG, etc. which work on any type
• Skips arguments that are already Cast/TryCast expressions (no double-wrapping)

Example

-- Input (Snowflake SQL)
SELECT SUM(amount) FROM t

-- After transform (DuckDB-compatible)
SELECT SUM(TRY_CAST(amount AS DOUBLE)) FROM t

Tests

• Unit tests in test_transforms.py covering SUM, AVG, COUNT (unchanged), MAX (unchanged), already-cast args, and mixed queries
• Integration test in test_fakes.py verifying end-to-end SUM/AVG on a VARCHAR column

[tekumara]

thanks for the contribution!