Add per-model token counts to cost history

[iam-brain]

Summary

Add end-to-end per-model token-count plumbing so the cost history chart can show real scanned token counts for each model.

What changed

• Added totalTokens to CostUsageDailyReport.ModelBreakdown and preserved it during decoding.
• Updated Codex and Claude scanners to carry per-model token totals into each day's modelBreakdowns.
• Removed scanner-side truncation so the full deterministic breakdown reaches the UI.
• Sorted model breakdowns deterministically by cost descending, then token count descending, then model name descending.
• Kept the row cap in CostHistoryChartMenuView and updated row detail rendering to show cost and token count when available.
• Added totalTokens to the CLI modelBreakdowns payload for shape consistency.
• Split scanner breakdown coverage into a focused test file instead of growing the existing scanner test type.

Verification

• pnpm check
• swift test
• ./Scripts/compile_and_run.sh