- Read schema from SQL CREATE TABLE scripts instead of inferring from JSON
- Stream JSON records using Utf8JsonReader instead of loading all into memory
- Write protobuf output in batches of 10000 rows to reduce memory usage
- Add mapping from cache file names to SQL scripts and table names
- Map SQL types (VARCHAR, BIGINT, DECIMAL, DATETIME2, BIT) to .NET types
- Update usage to require scripts directory as second argument