[BUG] How should the consumed tokens be counted when the inference is interrupted halfway or canceled

### 📋 CheckList

- [x] I have searched existing issues to avoid duplicates
- [x] I am using a recently maintained version of Coze Loop
- [x] I have provided all required information
- [x] I understand this is a bug report and not a feature request
- [x] I have submitted this report in English (otherwise it will not be processed)

### 🐛 Bug Description

How should the consumed tokens be counted when the inference is interrupted halfway or canceled
The current statistics are based on the last output of usage_age. If the inference fails halfway, how can it be completed.

### 🔄 Steps to Reproduce

Do it by yourself !