read_json with orient="values" and numpy=True provides strange column

Another edge case revealed in #28510 . Perhaps won't be relvant if we go through with #28512 but opening for posterity

For whatever reason, this particular combination of values produces a RangeIndex for the columns.

```python
>>> pd.read_json("[]", orient="values", numpy=True).columns
RangeIndex(start=0, stop=1, step=1)
```

Nothing else appears to do that