[benchmark] Add DataFusion TPC-DS benchmark harness

[JingsongLi]

What changed

• Add a non-published benchmarks/tpcds workspace crate with load and run commands.
• Load external TPC-DS-generated Parquet data into append-only Paimon tables.
• Run the same external query files against Paimon or native Parquet through DataFusion.
• Record planning, execution, row-count, spill, and available physical metrics in a versioned JSON report.
• Add configurable SessionConfig and RuntimeEnv constructors to SQLContext for benchmark memory, parallelism, and spill settings.
• Document SF10/SF100/SF1000 operation, cache protocol, and result interpretation.

Why

The DataFusion integration has focused functional tests but no reproducible macro benchmark for analytical joins, aggregations, sorts, and spill behavior. This harness provides an end-to-end comparison using the same binary, runtime settings, and external query files for both sources.

The repository does not include TPC tools, generated data, or query text. Reports are explicitly labeled as TPC-DS-derived non-TPC results.

Validation

• cargo build -p paimon-tpcds-bench
• cargo test -p paimon-tpcds-bench — 7 unit tests and 4 smoke tests passed
• cargo test -p paimon-datafusion --test sql_context_tests — 48 tests passed
• cargo clippy -p paimon-tpcds-bench -p paimon-datafusion --all-targets -- -D warnings
• cargo fmt --all -- --check
• git diff --check

[leaves12138]

Reviewed the latest changes locally. The benchmark tests, targeted SQLContext test, formatting, clippy checks, and the real Parquet load/overwrite smoke verification all passed. The remaining TPC-DS SQL compatibility work is intentionally handled in a follow-up PR.