Display prefix cache hit token statistics in LLM Metrics summary

[Pernekhan]

Prefix caching is a key factor in LLM inference performance, but there's currently no visibility into cache behavior in the aiperf output.

[jeongyoonm]

@ajcasagrande Hello! Wondering if this is going to be implemented soon. Otherwise, I would like to implement this feature for SGLang (and other inference engines if possible).