[massimim]

Introduces a new single-node strong-scaling benchmark orchestrator and its accompanying documentation. This enhancement streamlines performance evaluation across various Carbon benchmarks on a single machine.

• Facilitates strong-scaling analysis: A new orchestrator script enables automated sweeps over GPU counts for benchmarks like Jacobi, LBM, micro-benchmarks, and halo, providing consistent performance insights.
• Enables LBM multi-GPU control: The LBM benchmark now accepts a --gpus N command-line argument, allowing it to dynamically limit the number of CUDA devices utilized, which is crucial for participation in scaling experiments.
• Provides clear usage guidelines: Comprehensive documentation details how to build, configure, run, and interpret the results of these strong-scaling benchmarks.