Option for JSON parser to return NULL field values on type mismatch

[scovich]

Is your feature request related to a problem or challenge? Please describe what you are trying to do.

JSON is notoriously semistructured, but as far as I can tell arrow-json's ReaderBuilder only allows a very strict parsing of JSON rows to a specific homogenous schema. There is an option to ignore unwanted columns (with_strict_mode(false)), and for converting boolean/numeric values to strings (with_coerce_primitive(true)), but any other kind of schema mismatch on even a single row produces a hard error for the whole batch.

Describe the solution you'd like

It would be nice to have some ReaderBuilder option that converted wrong-type values to NULL instead of forcing a parsing error. For example, Spark's (woefully underdocumented) from_json function does this by default.

NOTE: This is NOT a request to handle malformed JSON. If it can't even json-parse it should be an error. This is about handling the case where e.g. the schema requested an int and the JSON provides an array of ints.

Schema disagreements like this frequently come up due to versioning issues or disagreements between different parties exchanging data in JSON format; in such cases, getting partial data is arguably better than getting nothing at all.

Is this something that could be considered desirable to support?

Describe alternatives you've considered

One alternative might be to request a schema with all-string leaf fields and manually parse values afterward. Spark's from_json has such an option. But that doesn't handle the case where the data provides an array or object where the schema expected a leaf, and it also doesn't handle the case where the schema expected a non-leaf column like a struct and got an array or primitive instead.

Additional context

Some examples where it would be helpful to tolerate partially incompatible schemas:
delta-io/delta#2419
delta-io/delta-kernel-rs#501
delta-io/delta-kernel-rs#712