feat: DLQ support, error classification, and retry tracking

[jamesdube]

Summary

• Add error classification types (PermanentError, TransientError) with IsPermanent()/IsTransient() helpers for consumer error routing
• Add DLQ infrastructure: dead-letter exchange ({exchange}.dlx) + DLQ queue ({queue}.dlq) auto-declared on consumer setup
• Add ConsumerConfig struct and NewConsumerWithOptions() constructor for full control over retry/DLQ/prefetch behaviour
• Implement retry tracking via x-retry-count AMQP header with exponential backoff (capped at 30s)
• Add panic recovery in handler — panics become PermanentError and route to DLQ
• Fix Ack(true)/Nack(true, ...) multiple=true bug that could ACK/NACK unprocessed messages when prefetch > 1

Error routing logic

Error type	Behaviour
PermanentError	Route to DLQ immediately, ACK original
TransientError / unclassified (retries < max)	Republish with incremented x-retry-count, exponential backoff, ACK original
Unclassified (retries >= max)	Route to DLQ, ACK original
Handler panic	Wrapped as PermanentError → DLQ

Backward compatibility

• NewConsumer() and NewConsumerWithConfig() unchanged — DLQ disabled by default
• Existing consumers get Nack(false, true) (requeue) on error, same as before but with the multiple=false fix

Test plan

• 12 new unit tests for error types, header parsing, config defaults, and header copying
• All existing tests pass with -race
• Integration test with live RabbitMQ to verify DLQ routing end-to-end

🤖 Generated with Claude Code