datafusion-functions-json functions parity

[sdf-jkl]

List of functions to implement:

[friendlymatthew]

Very very very very VERY nice!

[sdf-jkl]

@friendlymatthew I found parity issues in the current logic for some of the functions.

This is a convenience wrapper equivalent to:
variant_get(variant, path, 'Utf8')

Behavior (null/error/casting) should match variant_get with a type hint.

Since we're going for parity with both: the main variant_get function and datafusion-functions-json there are conflicts:

• variant_get_str currently stringifies non-string values. That diverges from both variant_get(..., 'Utf8') and json_get_str, which return string values and otherwise return Null. Easy fix.
• variant_get_float currently coerces Int to Float to match json_get_float. However, this diverges from variant_get(..., ('Float64'), which does not coerce all integers to float. If we have to pick, we should go for variant_get parity.

Codex drew a parity table for me:

JSON func	Closest upstream variant_get equivalent	Parity	Key differences
json_get	variant_get(..., as_type=None)	Partial	Both do path extraction + null on missing path, but return types differ: JSON returns a JSON union, upstream returns VariantArray.
json_get_str	variant_get(..., as_type='Utf8')	High	Both are string-only; non-string becomes null.
json_get_int	variant_get(..., as_type='Int64')	High	Both are int-only; float/non-int become null.
json_get_float	variant_get(..., as_type='Float64')	Low	JSON coerces int->float (int.into()), upstream float extraction is as_f64() only.
json_get_bool	variant_get(..., as_type='Boolean')	High	Both are bool-only; non-bool becomes null.
json_get_json	variant_to_json(variant_get(..., as_type=None))	Partial	JSON returns raw source slice text; variant path + re-serialization can normalize representation (e.g. exponent form).
json_get_array	No direct single-call equivalent	Low	JSON returns List of each element’s JSON text. Upstream list casting is typed (List), not “JSON text per element”.