feat: populate UsageUpdate.cost with estimated session cost

[simonrosenberg]

Summary

Populate UsageUpdate.cost with an estimated cumulative session cost derived from per-turn token counts and a model pricing table.

• Add pricing.rs with per-model pricing for gpt-4.1, gpt-4o, o3, o4-mini families and prefix-matching for date-suffixed model slugs
• Track cumulative cost across the session via Rc<Cell<f64>> shared between ThreadActor and PromptState
• Compute incremental cost from last_token_usage on each TokenCountEvent
• Update model tracking on ModelReroute events so mid-session model switches are priced correctly
• For unknown models, cost is omitted (None)

This is an interim solution until the upstream Codex event stream exposes authoritative cost (openai/codex#16258).

Caveats (documented in #209):

• Service tier / fast mode pricing is not accounted for (no upstream signal — see openai/codex#13794)
• Pricing table needs periodic updates as OpenAI changes rates
• ChatGPT subscription users see estimated API-equivalent cost (not their actual billing)

Closes #209

Test plan

• CI build passes
• Verify UsageUpdate notifications include cost with amount > 0 and currency: "USD" for known models
• Verify cost is omitted for unknown models
• Verify cost accumulates correctly across multiple turns
• Verify model reroute updates pricing for subsequent turns
• Unit tests in pricing.rs pass

🤖 Generated with Claude Code