测试自定义词典分词效果较差

case示例：
```
from LAC import LAC
l = LAC(mode='seg')
l.add_word("广东省/n")
l.add_word("人/n")
l.add_word("民/n")
l.add_word("人民/n")
l.add_word("民政/n")
l.add_word("政府/n")
```
```
l.run("人民政府")
['人', '民', '政府']
```
期望得到 ['人民', '政府']
我观察在源码 `parse_customization` 中仅根据前缀树来修正，有办法引入概率或者其他方式来解决优化吗？

Reproduce the reported case through the Python LAC API and inspect the parse_customization path that currently applies prefix-tree corrections. Compare the output for the registered words with the expected ['人民', '政府']; done means custom dictionary segmentation handles this case without regressing other segmentation behavior.

Written by the indexing model from the issue text.