In order to use a DKIM public key, it has to be published in the DNS system.
The most easy setup simply puts it directly into the appropriate TXT record.
A more advanced setup also allows them to be delegated to some other record
pointed to by a CNAME record.

As an example, both variants are equivalent in terms of DKIM key validation:

selector1._domainkey.example.org IN TXT "v=DKIM1; p=..."

and

selector1._domainkey.example.org IN CNAME "somekey.mailprovider.org."
somekey.mailprovider.org IN TXT "v=DKIM1; p=..."

So, the actual key management can be delegated to some other party by means of
using a CNAME.

This is of special importance for infrastructure automation and DKIM key rotation
(we just ask the customer / domain dns owner once to add the appropriate CNAME
pointing to the corresponding record on our own domain, and then we're able to
modify the DKIM keys on our own without requiring the customers to update their
DNS records all the time).
As key rotation is a current best practice, the added CNAME support will ease
the respective implementation.

This commit changes the code to first check a TXT record (as before), and in
addition checks a CNAME record (and if found, check the record it is referring
to - up to a limit of 10 recursion steps to avoid endless recursions).