Test harness fidelity: consumer retry, Fault<T>, and request timeout

[Vulthil]

What

Closes the Vulthil.Messaging.TestHarness fidelity gaps so behaviour verified against the in-memory harness matches real RabbitMQ.

TestHarness

• Consumer retry — the one-way handler now applies the consumer's retry policy, looping attempts in a fresh DI scope each time (mirroring RabbitMqConsumerWorker, without the real back-off delays). A throwing consumer no longer propagates to the publisher.
• Fault<T> emission — on retry exhaustion a Fault<TMessage> is published (captured and delivered in-process), so harness.Published<Fault<T>>() works.
• Request timeout — a request with no consumer/responder now returns Messaging.Request.Timeout (matching the real requester), not Error.NotFound.
• RetryCount / Redelivered — populated from the attempt number instead of hardcoded 0/false.

Behaviour changes (docs + tests updated): a one-way consumer that throws lets the publish/send succeed and emits a fault; a no-consumer request times out.

Messaging-core cleanups (first commit)

• Removed the redundant IMessageConfigurationProvider.GetMessageConfiguration<T>() overload.
• MessageContext.CorrelationId is now nullable, honouring the IMessageContext contract (dropped the ?? string.Empty coercion).

Tests

3 new TestHarness tests (fault emission, retry-with-incrementing-RetryCount recovery, no-consumer timeout); full TestHarness, core-messaging, and RabbitMq suites green locally (net9.0 + net10.0).

🤖 Generated with Claude Code