bigquery: Multi-stream ReadSession acceleration for large query results #7039

Extend StorageReader to automatically upgrade large query result reads (above the multi-stream row threshold) to a parallel multi-stream ReadSession against the query job's destination table.

• Fetch the query job's destination_table via JobService::get_job when total_rows / total_estimated_row_count exceeds the multi-stream threshold.
• Create a ReadSession with DataFormat::Arrow and CompressionCodec::Lz4Frame on the destination table.
• Spawn concurrent stream reader tasks across session.streams sending decoded RecordBatches over a bounded channel to StorageReader::next_batch() (ensuring proper backpressure and task cancellation on drop).
• Unit and integration tests for multi-stream session creation and fallback behavior when no destination table or only a single stream is available.