Before, 1-2k record took ~10s and more than that was just painful, 5k took
forever. This records things to keep a dict of nodes with a set of records so
that we can quickly "jump" to the point we're looking for without having to
search. 10k records now takes ~5s.

Support replace=True in zone.add_record

- unrecognized wording for invalid values

- punting on service, regex & replacement validation for now
- clean up MX a smidge

- adds lenient flag to Record.new, problems during validation are just
  warnings if it's true
- target populate calls during the plan phase pass lenient=True
- make all of the provider.populate call logging consistent including both
  target and lenient
- add source=self to Record.new in a few places that were missing it

This is with an eye toward expanding it in the future both in terms of what it
checks and to add the ability to ignore things. This commit does not intend to
change any validation. It only reworks the flow and improves the error
messaging.

Thanks @offmindby!

```yaml
ignored:
  octodns:
    ignored: true
  type: A
  value: 1.2.3.4
```

They were working around the lack of class hierarchy, this addresses that by
adding 2 child classes. It gets rid of a bunch of ugly default params and
if-this junk in the main class that was trying to deal with plain & geo records.
Also as a positive side effect it gets rid of some hacks that lived in
Route53Provider dealing with the fact that the old setup couldn't detect going
to/from a GeoDNS record correctly.