0-length valued arrays fail to save if array_inline_threshold is set

```python import asdf, numpy as np cfg = asdf.get_config() cfg.array_inline_threshold = 1000 tree = {"scalar": np.zeros([], dtype="float32")} asdf.dumps(tree) ``` produces a validation error: ``` 'id': 'http://stsci.edu/schemas/asdf/core/ndarray-1.1.0', 'title': 'An *n*-dimensional array.\n'} On instance: {'shape': [], 'data': 0.0, 'datatype': 'float32'} ``` These are honestly pretty odd arrays since they have no shape but a value. These do currently round-trip via internal storage (and are being created in roman_datamodels test data). However I'm not sure if the schema currently allows saving them inline. It would be good to figure out what options we have here.