add a to-json subcommand

[thinkingfish]

Claude did most of it. My main question is if there is a way to make the type/array conversions less verbose than the existing method, if not, then the general code quality seems fine.

[mihirn]

I think the cleaner way of doing type conversions (at least in one's own code, I suspect it does something similar in the library) is to use the arrow_json::writer (https://docs.rs/arrow-json/57.1.0/arrow_json/writer/struct.Writer.html) on the RecordBatch one is iterating over.

I think one can filter the columns either during loading using a ProjectionMask (https://arrow.apache.org/rust/parquet/arrow/struct.ProjectionMask.html) or after loading using filter_record_batch (https://arrow.apache.org/rust/arrow_select/filter/fn.filter_record_batch.html), but I haven't tried any of them myself.

If we don't want to experiment with either of those, I don't know a cleaner way, and this is fine (though I suspect it will baulk with larger files since it is keeping all the data in memory).