Run RabbitMQ deliveries through the core DeliveryDispatcher, with one scope per consumer attempt and no reply on shutdown

[Vulthil]

Summary

Moves the RabbitMQ transport onto the core DeliveryDispatcher from #422. The worker keeps only the AMQP parts: ack,
nack for dead-lettering, the retry-queue re-publish, the fault route and the reply route.

Production behavior changes

• Every consumer attempt runs in its own DI scope. Before, all consumers of one delivery shared one scope per round,
  so a failed consumer's scoped state (for example an unsaved DbContext) could reach the next consumer.
• A request consumer stopped by shutdown sends no reply, and the delivery stays unsettled so the broker delivers it
  again. Before, it replied with an RpcFault ("A task was canceled").
• A fault's OriginalContext is the context of the attempt that failed for good, so its RetryCount is the round the
  consumer failed on. For a one-way envelope delivery its RequestId is now empty; it held the AMQP correlation id.
  The fault's Message is still the payload as delivered.
• Logging: the RabbitMQ events 1103–1105 are removed, and the dispatcher's 2200–2202 replace them. Each delivery opens
  a logging scope with the queue, the routing key and the message type, so those entries keep that context.
• Tracing: every failed attempt is recorded as an exception event on the delivery's receive span.

Fix

• WithRetryCount and the retry re-publish copied the delivery's AMQP properties, but the copy shared the header
  dictionary with the delivery, so setting x-retry-count also changed the original delivery. Both now copy the
  headers.

Code

• MessageHandler, RabbitMqHandlerFactory and GatedPublisher are deleted. QueueDispatchPlans builds
  DeliveryHandlers with DeliveryHandlerFactory.
• RabbitMqBus takes the DeliveryDispatcher from DI instead of an IServiceScopeFactory.
• No public API change: every changed type is internal.

Tests

• ConsumeFilterPipelineTests, PolymorphicDispatchTests and RecordingGatedPublisher are deleted, because they
  tested pipeline and dispatch rules the core now owns. A core test pins that a request's filters finish before its
  reply is sent, and a plan test pins which consumers a polymorphic plan holds.
• The worker retry tests keep the AMQP specifics: retry queue, TTL, headers, ack/nack, in-process wait and shutdown.
• New worker tests: the reply route and correlation id, no reply on shutdown, the logging scope reaching the
  dispatcher's entries, the fault round, and WithRetryCount leaving the delivery unchanged.

Docs

• Retries: the scope rule and the shutdown rule. Faults: OriginalContext is the failing attempt. Request/Reply: no
  reply on shutdown. Observability: the dispatcher's log events, the delivery scope and the span events.

Verification

• Full solution build: 0 warnings, 0 errors.
• All test projects pass on net10.0 and net9.0. This includes Vulthil.IntegrationTests (Testcontainers with a real
  broker, 68 tests) and the Aspire messaging integration tests (net10.0, 13 tests).
• Mutation checks:
  • Publishing the fault with the old delivery snapshot fails the fault-round test.
  • Opening the delivery scope on another logger fails the log-scope test.
• dotnet pack of Vulthil.Messaging.RabbitMq passes package validation against the 1.2.0 baseline.

Backport to v1.0: no

[claude]

Code review

No issues found. Checked for bugs and CLAUDE.md compliance.