Implement customizable task dispatch interceptor scheme

[conradbzura]

Summary

Implement a customizable interceptor system for task dispatch that allows users to hook into the task lifecycle — modifying tasks before dispatch and wrapping response streams during execution. Interceptors follow an async generator protocol with three phases: pre-dispatch, stream processing, and cleanup.

Add an InterceptorLike protocol, an @interceptor decorator for global registration, an InterceptorBridge that adapts Wool interceptors to gRPC's server interceptor interface, and integrate interceptor support into LocalWorker and WorkerProcess.

Motivation

Cross-cutting concerns such as logging, metrics, RBAC, and distributed tracing need a clean extension point in the task dispatch path. Without an interceptor mechanism, users must modify worker internals or wrap proxy calls ad hoc. A first-class interceptor protocol provides a composable, testable, and well-documented way to observe and transform tasks and their response streams.

Affected code

• wool/src/wool/runtime/work/interceptor.py — new module
• wool/src/wool/runtime/worker/local.py — interceptor integration
• wool/src/wool/runtime/worker/process.py — interceptor bridge wiring
• wool/src/wool/runtime/worker/connection.py — type annotation fix
• wool/src/wool/runtime/loadbalancer/roundrobin.py — type annotation fix
• wool/src/wool/runtime/work/task.py — type annotation fix, assertions
• wool/src/wool/__init__.py — public API exports
• wool/pyproject.toml — grpc-interceptor dependency
• wool/tests/runtime/work/test_interceptor.py — comprehensive test suite