[datafusion] Align Paimon scan statistics with Parquet

[JingsongLi]

Summary

Align Paimon scan statistics with DataFusion native Parquet semantics. This exposes conservative manifest-derived column statistics while avoiding compressed full-file bytes that do not represent projected scan bytes.

Changes

• decode per-file value statistics using the current schema and dense column mappings
• aggregate manifest min/max bounds and publish exact null counts only for row-preserving scans
• fail open on malformed, inconsistent, schema-mismatched, or ordering-incompatible statistics
• suppress unsafe statistics for aggregation merge engines, filters, limits, row ranges, and deletion files
• report total_byte_size as unknown because Paimon only stores compressed full-file sizes
• add regression coverage for aggregation, unsafe null-count paths, malformed rows, and binary ordering

Testing

• cargo test -p paimon --all-targets --features fulltext,vortex
• cargo test -p paimon value_stats_for_field
• cargo test -p paimon-datafusion --lib test_partition_statistics
• cargo test -p paimon-datafusion --lib test_binary_manifest_bounds
• cargo clippy --all-targets --workspace --features fulltext,vortex -- -D warnings
• cargo fmt --all -- --check
• git diff --check

Runtime filter support remains a separate follow-up.

[leaves12138]

Looks good to me. The manifest-derived bounds are conservative, exact null counts are restricted to row-preserving scans, and malformed or schema-mismatched statistics safely fall back to unknown.