bigquery: benchmark arrow + jobs.query, Storage Read API and define routing thresholds #7041

This issue tracks benchmarking for BigQuery query result reading across:

1. Standard REST JSON pagination (jobs.query + jobs.getQueryResults)
2. Arrow results in jobs.query (QueryResultsFormat::Arrow, #[cfg(google_cloud_unstable_bigquery_arrow)])
3. Storage Read API acceleration (#[cfg(google_cloud_unstable_bigquery_storage_read)]) — both single-stream (jobs/{job_id}/streams/_default) and multi-stream (ReadSession on the destination table)

Goals:

• Measure performance & memory improvements:
  • Track end-to-end query time, row read/deserialization time, and throughput (rows/sec).
  • Track heap memory usage (stats_alloc: total bytes allocated, bytes/row, total allocation count, allocations/row) for both raw Row iteration and typed #[derive(FromRow)] deserialization.
• Determine routing thresholds:
  • Find the crossover threshold where switching from inline/paginated REST results to opening a Storage Read API stream becomes faster.
  • Evaluate whether and at what threshold the extra RPC overhead of jobs.get + CreateReadSession + parallel gRPC streams outperforms reading directly from the single _default job stream (projects/{p}/locations/{l}/jobs/{j}/streams/_default).