TypingAndDedupingTests: RecordDiffer should accept airbyte metadata column names instead of hardcoding them

[edgao]

minor tech debt from #30056

Snowflake's T+D expected records (in both sqlgenerator and TDTest) have lowercase _airbyte_extracted_at, etc. even though the columns are actually uppercased in the destination. This is because SnowflakeTestUtils now has a dumb hack to downcase those columns when dumping the final table.

Remove that hack. Update the expected records to have uppercase _AIRBYTE_* column names.

Update RecordDiffer to accept the metadata column names in the constructor, rather than hardcoding lowercased _airbyte_* everywhere. Then snowflake can pass in uppercased names.