feat: centralize cost breakdown calculations in pipeline

Extract token-type and per-model cost calculations from cmd/costs.go
into a dedicated pipeline.AggregateCostBreakdown() function. This
eliminates duplicate cost calculation logic between CLI and TUI.

New types:
- TokenTypeCosts: aggregate costs by input/output/cache types
- ModelCostBreakdown: per-model cost components

Benefits:
- Single source of truth for cost calculations
- Uses LookupPricingAt() for historical accuracy
- Both CLI and TUI now share the same cost logic
- Easier to maintain and extend

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

internal/pipeline/costs.go

@@ -0,0 +1,93 @@
+package pipeline
+
+import (
+	"sort"
+	"time"
+
+	"github.com/theirongolddev/cburn/internal/config"
+	"github.com/theirongolddev/cburn/internal/model"
+)
+
+// TokenTypeCosts holds aggregate costs split by token type.
+type TokenTypeCosts struct {
+	InputCost     float64
+	OutputCost    float64
+	Cache5mCost   float64
+	Cache1hCost   float64
+	CacheReadCost float64
+	CacheCost     float64
+	TotalCost     float64
+}
+
+// ModelCostBreakdown holds cost components for one model.
+type ModelCostBreakdown struct {
+	Model         string
+	InputCost     float64
+	OutputCost    float64
+	Cache5mCost   float64
+	Cache1hCost   float64
+	CacheReadCost float64
+	CacheCost     float64
+	TotalCost     float64
+}
+
+// AggregateCostBreakdown computes token-type and model cost splits.
+// Pricing is resolved at each session timestamp.
+func AggregateCostBreakdown(
+	sessions []model.SessionStats,
+	since time.Time,
+	until time.Time,
+) (TokenTypeCosts, []ModelCostBreakdown) {
+	filtered := FilterByTime(sessions, since, until)
+
+	var totals TokenTypeCosts
+	byModel := make(map[string]*ModelCostBreakdown)
+
+	for _, s := range filtered {
+		for modelName, usage := range s.Models {
+			pricing, ok := config.LookupPricingAt(modelName, s.StartTime)
+			if !ok {
+				continue
+			}
+
+			inputCost := float64(usage.InputTokens) * pricing.InputPerMTok / 1_000_000
+			outputCost := float64(usage.OutputTokens) * pricing.OutputPerMTok / 1_000_000
+			cache5mCost := float64(usage.CacheCreation5mTokens) * pricing.CacheWrite5mPerMTok / 1_000_000
+			cache1hCost := float64(usage.CacheCreation1hTokens) * pricing.CacheWrite1hPerMTok / 1_000_000
+			cacheReadCost := float64(usage.CacheReadTokens) * pricing.CacheReadPerMTok / 1_000_000
+
+			totals.InputCost += inputCost
+			totals.OutputCost += outputCost
+			totals.Cache5mCost += cache5mCost
+			totals.Cache1hCost += cache1hCost
+			totals.CacheReadCost += cacheReadCost
+
+			row, exists := byModel[modelName]
+			if !exists {
+				row = &ModelCostBreakdown{Model: modelName}
+				byModel[modelName] = row
+			}
+			row.InputCost += inputCost
+			row.OutputCost += outputCost
+			row.Cache5mCost += cache5mCost
+			row.Cache1hCost += cache1hCost
+			row.CacheReadCost += cacheReadCost
+		}
+	}
+
+	totals.CacheCost = totals.Cache5mCost + totals.Cache1hCost + totals.CacheReadCost
+	totals.TotalCost = totals.InputCost + totals.OutputCost + totals.CacheCost
+
+	modelRows := make([]ModelCostBreakdown, 0, len(byModel))
+	for _, row := range byModel {
+		row.CacheCost = row.Cache5mCost + row.Cache1hCost + row.CacheReadCost
+		row.TotalCost = row.InputCost + row.OutputCost + row.CacheCost
+		modelRows = append(modelRows, *row)
+	}
+
+	sort.Slice(modelRows, func(i, j int) bool {
+		return modelRows[i].TotalCost > modelRows[j].TotalCost
+	})
+
+	return totals, modelRows
+}