[LichKing-2234]

Summary

• Add account-scoped, five-minute local prompt-cache estimates for OpenAI Chat Completions and Responses API usage details.
• Return standard cached_tokens and cache_write_tokens fields when a stable translated Kiro prefix meets the existing cache threshold.
• Reuse the existing local prompt-cache tracker model: record only after upstream success; use cumulative history-prefix breakpoints so growing conversations can reuse earlier prefixes.

Semantics

• These are local estimates, matching the existing Claude-side tracker approach; they are not Kiro upstream cache or billing telemetry.
• Estimates are isolated by Kiro account and exclude the current user message, current images, and current tool-result bodies.
• No Kiro request payload, account routing, session affinity, or billing behavior changes.

Validation

• go test ./...