Add an opt-in rfc2308_ttl mode for TTL inheritance

[peterthomassen]

Folks complaining about this:

Pre-RFC 2308 master files overloaded the SOA MINIMUM field to mean "default TTL for the zone". This is what dnspython does: a record with no TTL takes the SOA minimum whenever no $TTL directive has been seen. RFC 2308 section 4 removed that meaning of MINIMUM, leaving RFC 1035 section 5.1 (which says that a record without TTL inherits the most recently stated one).

This PR adds an rfc2308_ttl parameter (default False) to dns.zone.from_text(), dns.zone.from_file(), dns.zonefile.read_rrsets(), and dns.zonefile.Reader, which selects the RFC 2308 interpretation. An explicit default ($TTL or default_ttl/force_ttl) still wins in either mode, and the SOA minimum is still used as a last resort when nothing else has stated a TTL at all, so a zone whose only TTL is the SOA minimum parses identically in both modes.

How it works: the reader now tracks whether the current default TTL comes from SOA MINIMUM (default_ttl_from_soa, cleared by $TTL and saved/restored across $INCLUDE alongside the other TTL state), and the three places that resolved an missing TTL now use a single _implicit_ttl() helper instead of each open-coding the precedence. With the flag off, the resolved TTL is unchanged in every reachable state, so existing behavior is preserved.

[rthalley]

Given how old RFC 2308 is, and that no one likely wants the old behavior, I wonder if we should look at this as a bug and make the default True, even though that will have a backwards compatibility impact, at least in theory.

[peterthomassen]

Yes, that's why I was hesitating. I guess the only reason to keep the current behavior really is backwards compatibility.

I wouldn't mind making the new one the default, but I guess that would require a 2.9 release and an additional warning in the release notes or something. Your call!

[rthalley]

2.9 is coming in a week or two, so my vote is to make the proper behavior the default and add a note to whatsnew.rst.