feat(playground): show per-message cost using models.dev pricing

[samuv]

Summary

Adds a per-message USD cost next to the existing token breakdown on assistant messages in the playground chat. Pricing comes from models.dev (USD per 1M tokens), fetched once and cached on disk for 24h in the main process.

• Inline: 100 → 50 = 150 • $ 0.0012 next to the existing token totals.
• Tooltip: appends a Cost section with Input, Cached, Output, Total breakdown.
• Cached input tokens use the model's cache_read rate when models.dev exposes it (e.g. Claude Sonnet 4.5, Haiku 4.5); otherwise treated as regular input.
• No pricing → no cost rendered. Local providers (ollama, lmstudio) and unknown models render exactly as they did before.
• Coverage: openai, anthropic, google, xai, and all 188 OpenRouter models with cost entries (slash-prefixed IDs already match).

How it works

• main/src/chat/pricing.ts fetches https://models.dev/api.json, caches { fetchedAt, data } to userData/models-dev-cache.json, refreshes in the background when older than 24h, and falls back to the last cached copy when offline.
• IPC handler chat:get-model-pricing returns the extracted Record<providerId, Record<modelId, ModelCost>> map.
• Renderer hook useModelPricing caches the map in React Query (staleTime: 24h).
• calculateCost(usage, pricing) is a pure function — easy to unit-test, no React, no DOM.

What deliberately does NOT change

• No new fields on MessageMetadata — cost is derived on render from existing totalUsage + model + the cached pricing map.
• No changes to streaming, IPC transport, or persisted thread data.
• No thread-level totals (per design decision).
• No special-casing for ollama / lmstudio — they naturally have no pricing entry and fall through to "no cost rendered."

Test plan

• pnpm run lint
• pnpm run type-check
• pnpm run test:nonInteractive (212 files, 2402 tests, all green)
• Manual: open playground, pick an OpenAI / Anthropic / OpenRouter model with a known price, send a message → confirm inline $x.xxxx and tooltip breakdown appear after the response completes.
• Manual: switch to Ollama or LM Studio → confirm the message renders exactly as before (no cost line).
• Manual: airplane-mode reload → cached pricing still works.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

Adds per-message USD cost estimation to Playground assistant messages by fetching model token pricing from models.dev in the main process (disk-cached for 24h), exposing it via IPC, caching it in the renderer via React Query, and computing a cost breakdown from existing token usage metrics.

Changes:

• Added main-process pricing fetch + 24h disk cache and an IPC endpoint (chat:get-model-pricing) to expose a provider/model pricing map.
• Added renderer React Query hook to retrieve pricing and a pure calculateCost/formatUsd utility with unit tests.
• Updated TokenUsage UI to render inline cost and a tooltip breakdown when pricing is available (and added component tests).

Reviewed changes

Copilot reviewed 11 out of 11 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
renderer/src/features/chat/lib/calculate-cost.ts	Pure cost calculation + USD formatting helpers.
renderer/src/features/chat/lib/tests/calculate-cost.test.ts	Unit tests for cost calculation and formatting.
renderer/src/features/chat/hooks/use-model-pricing.ts	React Query hook to fetch/cache pricing map from IPC.
renderer/src/features/chat/components/chat-message/token-usage.tsx	UI: inline cost display + tooltip cost breakdown using pricing map.
renderer/src/features/chat/components/chat-message/assistant-message.tsx	Passes model into TokenUsage so pricing lookup can be model-specific.
renderer/src/features/chat/components/chat-message/tests/token-usage.test.tsx	Component tests for cost rendering behavior.
preload/src/api/chat.ts	Adds getModelPricing() to the preload chat API + types.
main/src/ipc-handlers/chat/pricing.ts	Registers IPC handler for chat:get-model-pricing.
main/src/ipc-handlers/chat/index.ts	Wires pricing IPC handler registration into chat IPC registration.
main/src/ipc-handlers/chat/tests/index.test.ts	Updates registration test to include pricing handler.
main/src/chat/pricing.ts	Implements models.dev fetch, extraction, in-memory cache, and disk cache w/ TTL + offline fallback.