Reports the current session's raw token usage, read directly from the host's own session transcript/rollout (auto-detected: Claude Code or OpenAI Codex CLI). Returns raw counts only — context_tokens (input + cache_creation + cache_read) plus a full breakdown and session metadata (session_id, model, timestamp) — with no percentage or context-window detection. May be unavailable early in a session before the first model response; then it returns { available: false, reason } instead of failing.