cleans up the Optimization section a bit to remove redundant content

• removes the hardware specific sections (GPU/CPU) since many of the same optimizations in these docs already exist
• give Optimum it's own section in the docs for more visibility and combines GPU/CPU usage for it
• give assisted decoding it's own section in the docs for more visibility as well and it's also not really a decoding strategy like the other search/sampling methods are
• also removes the Optimizing inference doc since it also has redundant content in it and will be replaced by the newer overview in [docs] optimizations quickstart #42538