Read Iceberg v3 Puffin deletion vectors

[osipovartem]

Summary

• decode Iceberg v3 deletion-vector-v1 blobs with framing, CRC, roaring-directory, and cardinality validation
• add PUFFIN manifest file-format support
• load only active-table DV blob ranges with bounded concurrent object-store requests
• apply compact roaring vectors through a batch DataFusion predicate without materializing delete rows
• preserve the existing v2 position-delete path and tables without deletes

Performance

• one exact ranged GET per relevant DV blob; no whole-Puffin read
• query-pruned data files do not trigger DV reads
• one file-path lookup per record batch and one roaring membership check per row
• no ScalarValue conversion or (path, position) row expansion

Validation

• cargo +1.95.0 test -p iceberg-rust-spec deletion_vector --lib
• cargo +1.95.0 test -p datafusion_iceberg deletion_vector_filter --lib
• cargo +1.95.0 test -p datafusion_iceberg --test position_delete
• cargo +1.95.0 clippy -p iceberg-rust-spec -p datafusion_iceberg --lib --tests -- -D warnings
• live Snowflake/Horizon fixture: Snowflake created a managed Iceberg v3 MOR table, inserted 100,000 rows, and deleted IDs 2/50000/99999; Rustice returned the matching 99,997-row aggregate and affected rows
• Rustice EXPLAIN showed iceberg_deletion_vector_keep(__data_file_path, __iceberg_file_row_position), confirming the Snowflake-generated Puffin DV path

The offline integration test creates an Iceberg v3 table, commits a Puffin deletion vector for row positions 1 and 3, and verifies SQL returns only rows 10, 30, and 50.