test: add real backend performance benchmarks

[j6k4m8]

Summary

• benchmark high-degree traversal across NetworkX, DataFrame, and SQL
• benchmark grouped SQL node insertion and SQL bulk edge insertion
• benchmark incremental DataFrame edge mutation
• benchmark metadata-bearing dataframe ingestion across local backends
• keep workloads limited to 250-500 elements for fast pull request feedback

[codspeed-hq]

Merging this PR will not alter performance

✅ 8 untouched benchmarks
🆕 9 new benchmarks

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
🆕	test_dataframe_incremental_edge_insertion	N/A	374.6 ms	N/A
🆕	test_dataframe_ingestion_with_metadata[dataframe]	N/A	1 ms	N/A
🆕	test_dataframe_ingestion_with_metadata[networkx]	N/A	8.2 ms	N/A
🆕	test_dataframe_ingestion_with_metadata[sql]	N/A	33.4 ms	N/A
🆕	test_high_degree_neighbor_traversal[dataframe]	N/A	935.1 ms	N/A
🆕	test_high_degree_neighbor_traversal[networkx]	N/A	3.6 ms	N/A
🆕	test_high_degree_neighbor_traversal[sql]	N/A	198.8 ms	N/A
🆕	test_sql_bulk_edge_insertion	N/A	191.1 ms	N/A
🆕	test_sql_grouped_transaction_node_insertion	N/A	160.3 ms	N/A

---

_{Comparing test/real-performance-benchmarks-88 (2f8620c) with master (ca03a49)}