v0.1.0 — Search & Tool Token Tracking

What's new:
• New Search/Tool column in the round table — web-search results and tool calls are now counted per round (prompt ↑ · search/tool ⌘ · answer ↓)
• Refreshed platform reference data (2026-08): new default models; Kimi context 256K → 1M; ChatGPT web context defaults to 27K (per-plan limits 27K/54K/128K, shown in settings)
• Round-tool legend explaining the Search/Tool "-" cell, in all 10 languages

Fixed:
• ChatGPT no longer double-counts your prompt in the Search/Tool column (the search call embeds the prompt text; duplicates are now deduped)