Understanding Inference Scaling for LLMs: Bottlenecks, Trade-Offs, and Perf / hacker news