pd.NA `na_rep` truncated in to_csv

Just found this while messing around with how the new pd.NA sentinel would work with extension modules

```python
>>> pd.Series(["a", pd.NA, "c"]).to_csv(na_rep="ZZZZZ")
',0\n0,a\n1,ZZZZZ\n2,c\n'
>>> pd.Series(["a", pd.NA, "c"], dtype="string").to_csv(na_rep="ZZZZZ")
',0\n0,a\n1,ZZ\n2,c\n'
```

Note the latter is truncated to two letters

@jorisvandenbossche 