fix: correct deepseek pricing, gemini streaming tokens, and group-name logging

- Add promo discount system for deepseek-v4-pro (75% off until 2026-05-31)
- Rewrite StreamGemini to handle both SSE and JSON array response formats,
  fixing 0-token logging for gemini-3-flash and gemini-3-flash-preview
- Fall back to model group name for cost lookup when concrete model
  isnt in the registry (fixes $0 cost on deepseek-auto entries)
- Move registry lock before FindModel call to fix data race

internal/utils/registry.go

@@ -45,6 +45,24 @@ func FetchRegistry() (*models.ModelRegistry, error) {
 	return nil, fmt.Errorf("failed to fetch registry after 3 attempts: %w", lastErr)
 }
 
+// promoDiscount describes a temporary pricing discount applied on top of
+// the standard (list) price from the model registry.
+type promoDiscount struct {
+	Factor    float64   // multiplier applied after standard calculation (0.25 = 75% off)
+	ExpiresAt time.Time // discount ends at this time (UTC)
+}
+
+// promoDiscounts maps model IDs to active promotional discounts.
+// Sources:
+//   - DeepSeek v4 Pro: 75% off list pricing until 2026-05-31
+//     https://api-docs.deepseek.com/quick_start/pricing
+var promoDiscounts = map[string]promoDiscount{
+	"deepseek-v4-pro": {
+		Factor:    0.25,
+		ExpiresAt: time.Date(2026, 5, 31, 23, 59, 59, 0, time.UTC),
+	},
+}
+
 func CalculateCost(registry *models.ModelRegistry, modelID string, promptTokens, completionTokens, reasoningTokens, cacheRead, cacheWrite uint32) float64 {
 	meta := registry.FindModel(modelID)
 	if meta == nil || meta.Cost == nil {
@@ -72,5 +90,12 @@ func CalculateCost(registry *models.ModelRegistry, modelID string, promptTokens,
 		cost += float64(cacheWrite) * (*meta.Cost.CacheWrite) / 1000000.0
 	}
 
+	// Apply promotional discounts (e.g. DeepSeek 75% off until 2026-05-31).
+	if discount, ok := promoDiscounts[modelID]; ok {
+		if time.Now().UTC().Before(discount.ExpiresAt) {
+			cost *= discount.Factor
+		}
+	}
+
 	return cost
 }