remove .codecs.DefaultXXXX and split into 3x/4x impls [LUCENE-3613]

Currently there is a lot of shared codec implementations for things like
field infos, segment infos, stored fields, term vectors under the .codecs
package with the name Default\*

I think actually the .codecs package should be our abstract codec apis,
and the implementations should be under the respective codecs.

Also, lots of these implementations are still 'shared' across the 3.x and 4.x
formats, but shouldn't be: the 4.x versions should be "clean", and the 3.x
versions should have all the backwards cruft.

Additionally, the "PreFlexRW" should be writing this stuff in 3.x format,
but today it does not (it writes a 3.x index repackaged in a 4.x "container").

Really PreFlexRW needs to write indexes that are exactly like the 3.x format
(if you opened them up in lucene 3.x you could read them!). When this works
then we know that codecs will actually work for our backwards compatibility needs
and we know that we have good index backwards compatibility testing.

For starters we could just pick some 3.x version of lucene and wire PreFlexRW
to it... in the future maybe the writer should take a version param so it
can actually simulate different minor versions.

---
Migrated from [LUCENE-3613](https://issues.apache.org/jira/browse/LUCENE-3613) by Robert Muir (@rmuir), updated Jan 17 2012
Attachments: [LUCENE-3613_vectors.patch](https://apache.github.io/lucene-jira-archive/attachments/LUCENE-3613/LUCENE-3613_vectors.patch) (versions: 2), [LUCENE-3613.patch](https://apache.github.io/lucene-jira-archive/attachments/LUCENE-3613/LUCENE-3613.patch)

Start by examining the shared implementations in the .codecs package and the PreFlexRW entry point, then review the attached LUCENE-3613.patch and vectors patch. Done means implementations are separated for the 3.x and 4.x formats, and PreFlexRW writes an index that Lucene 3.x can read.

Written by the indexing model from the issue text.