[Proposal][Mailgun][webhooks] change event_types and reject_reason mappings

[costela]

Currently, it seems impossible to differentiate temporary from permanent failures by analyzing AnymailTrackingEvent created by the mailgun webhook.
A delivery which had a temporary failure due to - for instance - a broken connection, will generate a webhook payload with (among other fields):

{
  "event": "failed",
  "reason": "generic",
  "severity": "temporary"
}

This gets mapped to a AnymailTrackingEvent with event_type=BOUNCED and reject_reason=BOUNCED, which is hardly helpful. In particular, in the case of a failed connection, it feels actually misleading, IMHO, since there was no actual bounce from the receiving server.

Looking at anymail/webhooks/mailgun.py, I'd suggest doing something like the following in esp_to_anymail_event():

if event_type == EventType.BOUNCED and event_data['severity'] == 'temporary':
  event_type = EventType.DEFERRED

Also, changing the reject_reasons mapping from generic → BOUNCED to generic → OTHER. The rationale being: if Mailgun cannot interpret the error, we shouldn't assume it has any particular type.

Additionally, and mostly unrelated, we also see "reason": "greylisted" in our logs, which is currently not handled at all and might be mapped to BOUNCED?

Before working on a concrete PR, I thought I get some opinions on this. Thoughts?

[medmunds]

Those changes both seem good to me.

The (root level) "reason" field in the Mailgun payload doesn't seem to be documented anywhere, and it's not clear to me that Anymail's Mailgun reject_reasons mapping is really adding any value. Perhaps the normalized reject_reason should just be BOUNCED for bounce events, and OTHER for rejected events (unless someone can figure out how to map Mailgun's event["reject"]["reason"] text to Anymail's normalized reject reasons).

Since Mailgun's "severity" field is at least partially documented (under their failed Event Type), it makes sense to use it to identify DEFERRED events.

Does "reason": "greylisted" arrive with "severity": "temporary"? If so, that's another good argument for this change—greylisted should be DEFERRED.