Generalize EnableDeadLetterQueueRecovery to SQS and Azure Service Bus (#3103)

[jeremydmiller]

Closes #3103.

Summary

RabbitMQ already has a first-class native-DLQ → durable-storage bridge via EnableDeadLetterQueueRecovery(). This generalizes that feature — with the same syntax — to Amazon SQS and Azure Service Bus, so natively dead-lettered messages land in wolverine_dead_letters and become queryable/replayable through IDeadLetters and tools like CritterWatch, instead of being invisible in a broker DLQ.

opts.UseAmazonSqsTransport().EnableDeadLetterQueueRecovery();          // SQS
opts.UseAzureServiceBus(cnx).EnableDeadLetterQueueRecovery();          // Azure Service Bus
// optional explicit names:
opts.UseAmazonSqsTransport().EnableDeadLetterQueueRecovery("orders-dlq");

What it does

• SqsDeadLetterQueueListener — background service that drains the SQS dead letter queue(s) used by listeners (or explicit names) and writes each message into durable storage via MoveToDeadLetterStorageAsync. Original exception type/message come from the stamped failure headers.
• AzureServiceBusDeadLetterQueueListener — drains both the Wolverine-managed dead letter queue (where buffered/durable endpoints move failures) and the native $DeadLetterQueue sub-queues of every listening queue/subscription (inline endpoints + ASB-native TTL/max-delivery). Exception metadata comes from stamped headers or the native DeadLetterReason/DeadLetterErrorDescription.
• Wolverine.Transports.DeadLetterRecoveredException (core, shared) implements IDeadLetterExceptionInfo so the original exception type is recorded in durable storage rather than the recovery wrapper.
• Messages are only deleted/completed off the broker after being safely persisted, so a transient storage outage never loses a dead letter.

Implementation note: I verified empirically that Wolverine's buffered/durable Azure Service Bus endpoints route failures to the managed wolverine-dead-letter-queue queue (not the native sub-queue, which is only inline + ASB-native). The ASB listener therefore drains both kinds of source, which is why the no-arg overload "just works" across endpoint modes.

Documentation

New "Recovering Native Dead Letters to Durable Storage" sections on the SQS, Azure Service Bus, and (previously undocumented) RabbitMQ dead-letter-queue pages, cross-linked so the consistent syntax is discoverable from each transport.

Tests

• Per-transport registration tests (no infrastructure) — settings holder + hosted service registered exactly once; name handling.
• End-to-end recovery integration tests — failing handler → broker DLQ → recovery listener → durable storage → queryable. Verified locally against LocalStack + Postgres (SQS) and the Azure Service Bus emulator + Postgres (ASB).

Verification

• Full wolverine.slnx Release build clean (0 warnings / 0 errors).
• All 8 new tests green; SQS and ASB recovery validated end-to-end against real infrastructure.

Pairs with

#3104 (per-endpoint dead-letter-destination contract) — filed/implemented separately.

🤖 Generated with Claude Code