feat: implement Kafka retry queue with hybrid retry strategy

[bashkirian]

Add Kafka-based retry queue for payout orchestration with a hybrid retry strategy inspired by Hyperswitch:

• Provider-level retry: quick retry same provider (short delays)
• Global retry: fresh provider selection when all providers fail
• Tiered backoff with jitter to prevent thundering herd

Infrastructure:

• Add Redpanda to docker-compose
• Create payout-retry-provider, payout-retry-global, payout-retry-dlq topics

Core changes:

• Add kafka-go dependency
• Create internal/queue package with Kafka publisher/consumer
• Create internal/worker package with retry worker
• Add retrying state and retry count tracking to payouts
• Add migration for retry tracking columns

Configuration:

• Tiered delays: provider [1s, 5s, 15s], global [30s, 2m, 5m, 15m, 30m]
• Random jitter up to 20% to prevent thundering herd

Closes #62