Connector migrations and redesign

[mwylde]

This PR redesigns the connector system and migrates (almost all) connectors into the arrow world

Operator/connector redesign

Currently, all operator infra and the operators themselves are implemented in arroyo-worker. Meanwhile connector definitions, configs, and testing logic (which often duplicates code in the connector implementation) is in arroyo-connectors.
To instantiate operators/connectors, we have a giant match blog in engine.rs.

This was (sort of) necessary back in the macro days, but with traits we can be more flexible.

There are a few things that are undesirable about this situation:

• Connector logic is split between arroyo-connectors and aroryo-workers and some stuff is duplicated (like the logic to construct a redis or kafka client).
• Adding a connector requires changing core arroyo code, and can't for example be added in a separate crate
• We have to maintain a map of all connectors/operators names in order to construct them

To address these issues this PR reorganizes the code as follows:

• Adds a new crate, arroyo-operator that contains the common operator infra (ArrowOperator, SourceOperator, BaseOperator, etc.)
• arroyo-connectors now depends on arroyo-operator, and all connector implementation has moved into that package
• The ArrowOperatorConstructor was reworked to be able to be a trait object, which gives us a unified way to construct operators without needing the big match block

Longer term, we can look into using a crate like https://github.com/dtolnay/linkme to allow connectors to be registered at compile time, making it easy for advanced users to add custom connectors just be adding a new crate to the build.

Connector migrations

All connectors have now been migrated into arrow with the exception of the FileSystem/Delta sink and nexmark. These were excluded as they require significant work to migrate.