Add reduce primitive examples and performance testing

[jayshah1819]

Changes

• Add standalone reduce example (reduce_example.mjs and .html)
• Add performance testing for reduce (reduceperf.mjs and .html)

Features

• Minimal working reduce example with validation
• Performance plots for bandwidth and runtime
• UI controls for datatype (u32/i32/f32), binop (add/max/min), input sizes, and trials

Part of documentation : For each primitive (reduce/scan/sort) we need:

• Standalone example
• Performance testing with plots
• M3 screenshots
• Algorithm descriptions

[jowens]

Question: Is it beneficial for us to have a separate standalone reduce example when it's going to be 95% the same code as the standalone scan example? I think the only differences will be picking reduce not scan as a primitive (of course) and allocating a different-sized output array.

I'm thinking about the long-term maintenance aspects here.