Handle `ActiveStorage::PurgeJob` failures on `RawEmail#erase`

[garethrees]

Extracted from #9040, since it's not a blocking issue for initial release but it is something we'll need to resolve.

We're currently using purge_later to purge the file attachment of RawEmail because, as per the documentation, it's "the recommended way to purge blobs from a transaction".

Since we're interacting with an external service (S3 at present) this job may fail. It's unlikely, given that the jobs will be retried for several attempts with a backoff strategy, but it is still possible that these may at some point fail without actually completing the purge, leaving us holding data we thought we'd destroyed.

As per this discussion, we don't currently receive notifications for dead jobs. We periodically check this, but in the case of RawEmail#erase I think we actually want to get notified or otherwise ensuring that the file gets purged.

[garethrees]

I think the the best approach here is to add a global job death exception notification: https://github.com/sidekiq/sidekiq/wiki/Error-Handling#death-notification

[gbp]

Probably go with ActiveJob's retry_on handler, when passed a block it is called after the retry attempts have been used up.