Support top-level arrays when reading JSON files

### Is your feature request related to a problem or challenge?

Datafusion can't read JSONL/ND-JSON files where the rows are arrays rather than objects. e.g.:

```
[1, 2, 3]
[4, 5, 6]
[7, 8, 9]
```

### Describe the solution you'd like

It would be great if Datafusion could directly consume ND-JSON files containing top-level arrays, mapping the array to a single field with a customisable name, like data or items.

I notice that the underlying arrow library seems to support this concept via `datafusion::arrow::json::ReaderBuilder::new_with_field`, but this doesn't seem to be exposed by the JSON `FileFormat` that Datafusion exposes.

### Describe alternatives you've considered

At present I am working around the issue with a pre-processing step that wraps each line with a { "data": ... } wrapper, however this comes with a heavy performance penalty.

### Additional context

_No response_

Start by tracing DataFusion's JSON FileFormat and how it constructs the Arrow ReaderBuilder; compare that path with datafusion::arrow::json::ReaderBuilder::new_with_field. Done means ND-JSON lines containing top-level arrays can be read directly, with the array mapped to a configurable field name such as data or items.

Written by the indexing model from the issue text.