Add metadata-only truncate snapshots and ID-aligned historical reads

[osipovartem]

Summary

• Add metadata-only Iceberg TRUNCATE snapshots with an empty manifest list, including v3 row-lineage accounting and commit conflict checks.
• Let historical DataFusion scans use an explicit Iceberg read schema. Align renamed, reordered, and added nested STRUCT/LIST/MAP fields by Iceberg field ID in Arrow batches; a reused name with a different ID yields a typed null.
• Reject non-increasing snapshot sequence numbers in local catalog commits.

Performance

TRUNCATE neither scans nor rewrites previous Parquet files. Unchanged-schema reads retain the existing adapter fast path. Schema-evolution mappings are planned once per file and applied using Arrow arrays and shared buffers, without per-row ScalarValue conversion.

Verification

• cargo +1.99.0 test -p iceberg-rust --lib truncate_: 8 passed.
• cargo +1.99.0 test -p datafusion_iceberg --lib: 113 passed, 2 manual benchmarks ignored.
• Targeted Clippy passed with only three pre-existing Rust 1.99 lints excluded.
• Rustice v3 TRUNCATE/CHANGES end-to-end tests passed against the pinned final commit.
• Independent read-only review: approved.

Parquet-backed LIST/MAP schema-evolution scans remain a follow-up; the adapter has focused Arrow-batch tests.