Empty coarse-grained POS tags for number in the large Romanian model

Hi, I noticed that the coarse-grained POS tags for numbers in the large Romanian model (`ro_core_news_lg`) is empty rather than `X`. Is this the expected behavior?

## How to reproduce the behaviour
```
import spacy
nlp = spacy.load('ro_core_news_lg')
for token in nlp('2025'):
print(len(token.pos_), token.pos_)
print(len(token.tag_), token.tag_)
```

## Your Environment
* Operating System: Windows 11 x64
* Python Version Used: 3.11.9
* spaCy Version Used: 3.8.4