Add schema evolution support for tables with changing column definitions

[shefeek-jinnah]

Description

Enable querying DuckLake tables whose Parquet files have evolved schemas across snapshots.

Problem

The current implementation assumes all Parquet files for a table share an identical schema, which breaks when columns are added, removed, or renamed across snapshots.

Solution

This change unifies file schemas at query time:

• Uses the resolved snapshot schema as the canonical schema
• Pads missing columns in older files with typed NULLs
• Maps renamed columns using DuckLake column identifiers
• Ensures consistent projection across all files

Result

Queries can safely span multiple snapshots with schema drift

[zfarrell]

We should keep an eye on duckdb/ducklake#149

We might be able to implement something in this lib, and then share how we approached it with the duckdb folks.