[alex-rogachev]

Add deletion and manual retry for tasks in the deferred dead-letter queue.

• DeadTasks#delete(id_or_ids) deletes one or multiple tasks and returns the number of distinct IDs removed.
• DeadTask#delete deletes an individual task and returns whether it was removed.
• DeadTask#retry enqueues the task with its original arguments, a new ID, and reset attempts, then removes the dead record. It uses the normal enqueue middleware and flushes pending writes when supported.
• Tasks retried outside a running server are persisted for execution after restart, with a warning.

Retry follows at-least-once semantics: concurrent retries or failures after enqueue can produce duplicates, so tasks and enqueue middleware must be safe to repeat. Middleware that suppresses enqueue still allows the dead record to be removed.

Gotchas:

• Queue#schedule does nothing when Iodine is not running. A task enqueued from the console or a Rake task is stored and runs after the next server start. retry prints a warning in this case.
• The Disk backend gets an optional flush_tasks method. retry calls it to write the new task to disk at once, without waiting for the fsync timer.
• The Disk backend now fsyncs the storage directory after it creates a new WAL file, so the file entry survives a crash.