max-throughput@1.0.0

Publisher claims: "Use when running compute-heavy work: model training, fine-tuning, evaluation, benchmarks, simulations, data preprocessing, dataset generation, compilation, test suites, or any long-running batch job, or when a job is slower than expected and needs performance tuning. Detect available CPU/GPU/memory resources, parallelize aggressively, then profile the running job to find the true bottleneck (data pipeline, CPU decode, GPU compute, VRAM, I/O) and tune batch size, DataLoader workers, pre-encoding, and precision accordingly to minimize wall-clock runtime." Observed in sandbox: no network, touches nothing outside the project, writes no memory.