"Returns true if the specified attribute has been set by the user or by
a database load and is neither nil nor empty?"

Fixes #1613

This is a first implementation, EXPLAIN is highly
dependent on the database and I have made some
compromises.

On one hand, the method allows you to run the most
common EXPLAIN and that's it. If you want EXPLAIN
ANALYZE in PostgreSQL you need to do it by hand.

On the other hand, I've tried to construct a string
as close as possible to the ones built by the
respective shells. The rationale is that IMO the
user should feel at home with the output and
recognize it at first sight. Per database.

I don't know whether this implementation is going
to work well. Let's see whether people like it.

Reasons:

 * Markdown reads well as plain text, but can also be formatted.
 * It will make it easier for people to read on the web as Github
   formats the Markdown nicely.
 * It will encourage a level of consistency when people are writing
   CHANGELOG entries.

The script used to perform the conversion is at
https://gist.github.com/1339263

This is to avoid confusing newbies, and to be consistent with the fact
that other options like :foreign_key already allow a symbol or a string.

Preserve SELECT columns on the COUNT for finder_sql when possible

Building the conditions of a nested through association could
potentially modify the conditions of the through and/or source
association.

This is a Bad Thing.

If a record is removed from a has_many :through, all of the join records
relating to that record should also be removed from the through
association's target.

(Previously the records were removed in the database, but only one was
removed from the in-memory target array.)

Fixes #3425.

Fixes creating records in a through association with a polymorphic source
type.

This change reduces churn in the db/development_structure.sql file when using :sql as
active_record.schema_format, and makes comparing diffs much easier.

Test ensures the output SQL-statements are lexically ordered by version.

Removed here 0d0176c4

Yes, we're on Rails 3.1 now.

Postgres: Do not attempt to deallocate a statement if the connection is no longer active.