refactor: delegate catalog scans to static table provider

[alexanderbianchi]

What changes

Have IcebergCatalogTableProvider::scan() load the latest table and construct a per-scan IcebergStaticTableProvider, then delegate to its scan(). The catalog provider keeps ownership of catalog refresh and writes; the static provider owns the read-plan construction. This PR is based on main and does not add eager file planning or depend on #67.

The generated static provider uses the catalog provider's cached Arrow schema instead of re-deriving a schema from the freshly loaded table. This preserves the existing contract: after the table evolves, a provider created before the schema change still exposes and scans its original columns. No public constructor or wire format changes.

The existing schema-evolution test now also scans a projected column after the table changes, checking that projection indices still refer to the provider's original schema. Existing catalog insert and lazy-read tests continue to run.

This is a separate refactor from #67; because both change the catalog scan method, whichever merges second may need a rebase.

Validation

• cargo test -p iceberg-datafusion --locked --test integration_datafusion_test test_scan_after_schema_evolution_reads_provider_columns: pass.
• cargo fmt --all -- --check: pass.
• cargo clippy --workspace --locked --all-targets -- -D warnings: pass.
• cargo test --workspace --locked: pass.
• git diff --check: pass.

AI disclosure

Developed with AI assistance and validated locally before opening as a draft.