API prompt precaching speeds up first-token generation

API提示预缓存加速首令牌生成

Claude API prewarms prompt cache with the system prompt, skips output, then hits cache on the real request.

Why it matters: HKR-H/K/R all pass: this is a concrete Claude API latency mechanism, not a vague product tease. It clears featured, but it is a mid-weight inference update rather than a major model or capability release.