Support MsgPack format/serdes

[priestprophetking]

"MessagePack is an efficient binary serialization format. It lets you exchange data among multiple languages like JSON. But it's faster and smaller. Small integers are encoded into a single byte, and typical short strings require only one extra byte in addition to the strings themselves."

MessagePack is supported by over 50 programming languages and environments.

https://msgpack.org/index.html#languages

[rahulbsw]

I opened PR #1068 for this: #1068

Implementation summary:

• Adds a new msgpack Arroyo format with msgpack.into_unstructured_json, msgpack.timestamp_format, and msgpack.decimal_encoding options.
• Adds source deserialization for structured MsgPack map records and an unstructured JSON mode for arbitrary MsgPack values.
• Adds sink serialization to MsgPack, including native MsgPack binary output for Arrow binary fields.
• Reuses the existing Arrow JSON decoding/encoding path where practical so MsgPack behavior stays aligned with JSON schema handling.
• Wires Kafka schema validation, OpenAPI/client type exposure, and schema validation for unstructured JSON mode.

A couple of intentional constraints are called out in the PR:

• Structured records must be MsgPack maps with string keys.
• MsgPack extension values are rejected for now instead of lossy conversion.

Validation run locally includes cargo test -p arroyo-formats, cargo test -p arroyo-rpc msgpack, targeted connector/planner checks, cargo check -p arroyo-api with a temporary Postgres DB, OpenAPI generation, and web UI type/build checks.