Arrow json decoding

[mwylde]

This PR adds initial support for decoding from JSON into arrow, via a slightly modified version of arrow-json (https://github.com/ArroyoSystems/arrow-rs/tree/49.0.0/json) that adds support for raw json fields.

I've also started the process of moving sources away from StructDef to using native arrow datatypes. In that world, we encode raw json as an extension type with the name arroyo.json. The datafusion project is also thinking about how to support JSON fields (apache/datafusion#7845), so hopefully in the future we'll be able to leverage something more native.

This also includes support for specifying json types in columns:

create table (
  a INT,
  b JSON
)

which will cause b to generate a raw json schema.

There are a few limitations to this initial PR:

• Due to how arrow-json works, error handling is done at a batch level rather than the individual record level, which means that with BadData::Drop, an entire batch gets dropped for a single bad record
• Only timestamps in postgres-compatible formats are supported (a superset of RFC3339 / ISO8601); in particular unix epoch millis are not supported
• The json-schema generator hasn't been updated yet

It's likely that our approach here will evolve, but for now this PR should work for most cases.