fix(billing): surface quota saturation events for admin auditing

Thread int32 saturation clamps from tiered settlement and video task
recompute into the consume/task logs under admin_info, so oversized or
malformed billing inputs stay auditable. Clamp negative audio duration
before token conversion and gate the saturation UI markers on admin.

relay/channel/openai/audio.go

@@ -105,7 +105,7 @@ func OpenaiTTSHandler(c *gin.Context, resp *http.Response, info *relaycommon.Rel
 		} else if duration > 0 {
 			// 计算 token: ceil(duration) / 60.0 * 1000，即每分钟 1000 tokens。
 			// duration 解析自上游返回的音频元数据，饱和转换防止 int 回绕。
-			completionTokens := common.QuotaFromFloat(math.Round(math.Ceil(duration) / 60.0 * 1000))
+			completionTokens := common.QuotaRound(math.Ceil(duration) / 60.0 * 1000)
 			usage.CompletionTokens = completionTokens
 			usage.CompletionTokenDetails.AudioTokens = completionTokens
 		}

relay/common/relay_info.go

@@ -163,6 +163,11 @@ type RelayInfo struct {
 
 	PriceData types.PriceData
 
+	// QuotaClamp is set (non-nil) when a quota conversion saturated at the
+	// int32 bound (or NaN fallback) while computing this request's charge.
+	// It is surfaced onto the consume/task log's admin_info for auditing.
+	QuotaClamp *common.QuotaClamp
+
 	// TieredBillingSnapshot is a frozen snapshot of tiered billing rules
 	// captured at pre-consume time. Non-nil only when billing mode is "tiered_expr".
 	TieredBillingSnapshot *billingexpr.BillingSnapshot

relay/relay_task.go

@@ -202,7 +202,9 @@ func RelayTaskSubmit(c *gin.Context, info *relaycommon.RelayInfo) (*TaskSubmitRe
 				quotaWithRatios *= ra
 			}
 		}
-		info.PriceData.Quota = common.QuotaFromFloat(quotaWithRatios)
+		quota, clamp := common.QuotaFromFloatChecked(quotaWithRatios)
+		info.PriceData.Quota = quota
+		noteTaskQuotaClamp(info, clamp)
 	}
 
 	// 7. 预扣费（仅首次 — 重试时 info.Billing 已存在，跳过）
@@ -278,7 +280,21 @@ func recalcQuotaFromRatios(info *relaycommon.RelayInfo, ratios map[string]float6
 			result *= ra
 		}
 	}
-	return common.QuotaFromFloat(result)
+	quota, clamp := common.QuotaFromFloatChecked(result)
+	noteTaskQuotaClamp(info, clamp)
+	return quota
+}
+
+// noteTaskQuotaClamp records the first quota saturation event onto the task's
+// RelayInfo so LogTaskConsumption can surface it on the submit log's
+// admin_info. First non-nil clamp wins.
+func noteTaskQuotaClamp(info *relaycommon.RelayInfo, clamp *common.QuotaClamp) {
+	if clamp == nil || info == nil {
+		return
+	}
+	if info.QuotaClamp == nil {
+		info.QuotaClamp = clamp
+	}
 }
 
 var fetchRespBuilders = map[int]func(c *gin.Context) (respBody []byte, taskResp *dto.TaskError){