Reliant on https://github.com/rails/rails/pull/15747 but pulled to a
separate PR to reduce noise. `has_many :through` associations have the
undocumented behavior of automatically detecting counter caches.
However, the way in which it does so is inconsistent with counter caches
everywhere else, and doesn't actually work consistently.

As with normal `has_many` associations, the user should specify the
counter cache on the `belongs_to`, if they'd like it updated.

This allows using polymorphism for the uninitialized attributes raising
an exception behavior.

This will make it less painful to add additional properties, which
should persist across writes, such as `name`.

Conflicts:
	activerecord/lib/active_record/attribute_set.rb

[ci skip]

/cc @chancancode @zzak

Moved `Builder` to its own file, as it started looking very weird once I
added private methods to the `AttributeSet` class and the `Builder`
class started to grow.

Would like to refactor `fetch_value` to change to

```ruby
self[name].value(&block)
```

But that requires the attributes to know about their name, which they
currently do not.

This is a partial fix for #15853. It only works when a `preload`
is issued and not an `eager_load`. I've added a skipped failing
test-case to keep in mind that we need to deal with `eager_load`.

We previously only did this if the old value was zero, to make sure
numericality validations run and failed if the user gave 'wibble' as the
value, which would be type cast to 0. However, numericality validations
will fail if there are any non-numeric characters in the string, so 5 ->
'5wibble' should also be marked as changed.

We still had one file using `column_for_attribute` when it could return
nil, causing deprecation warnings in the tests.

Adding `# :nodoc:` to the parent `class` / `module` is not going
to ignore nested classes or modules.

There is a modifier `# :nodoc: all` but sadly the containing class
or module will continue to be in the docs.

/cc @sgrif

This is public API, and `simple_form` depends on the `nil` return value.
We need to go through a deprecation cycle to return a null object. If
people want hash access, they can access the hash.

/cc @sgrif

This was used more previously, but other uses have been removed.

There is a class method with the same name that is tested, but not the
instance method.

`reload` is meant to put a record in the same state it would be if you
were to do `Post.find(post.id)`. This means we should fully replace the
attributes hash.

Attempting to reduce the number of places that care about the details of
how type casting occurs. We remove the type casting of the primary key
in `JoinDependecy`, rather than encapsulating it. It was originally
added for consistency with
https://github.com/rails/rails/commit/40898c8c19fa04442fc5f8fb5daf3a8bdb9a1e03#diff-06059df8d3dee3101718fb2c01151ad0R211,
but that conditional was later removed in
https://github.com/rails/rails/commit/d7ddaa530fd1b94e22d745cbaf2e8a5a34ee9734.

What is important is that the same row twice will have the same value
for the primary key, which it will.

This has been around for a couple of versions now, a `NoMethodError`
should suffice at this point.

The :singleton-method: directive is printed in the docs.
This directive is not necessary as the method is documentable just fine.

Mostly delegation to start, but we can start moving a lot of behavior in
bulk to this object.

Now the following case will work fine

    class Tag < ActiveRecord::Base
    end

    class Publisher::Article < ActiveRecord::Base
      has_and_belongs_to_many :tags
    end

Fixes #15761

`Type::Integer.new.type_cast('') # => nil`, we do not need a special
case to handle this, `nil => ''` already returns false. The only case we
need to handle is `0 => 'wibble'` should be changed, while `0 => '0'`
should not.

We're never going to be able to use the attribute object here, however,
so let's just accept the ugly demeter violation here for now.

Remove test cases which were either redundant with other tests in the
file, or were actually testing the type objects (which are tested
elsewhere)

Moves towards removing type casting knowledge from the column entirely

Also takes a step towards supporting types which use a character other
than ',' for the delimiter (`box` is the only built in type for which
this is the case)