As we move towards removing the `klass` method from `column`, the else
clause as it stands right now will become the only case.

This reverts commit be1db994, reversing
changes made to 6680ee94.

Reason: This changed a behavior where numeric attributes when receiving
blank values didn't change its value.

[ci skip]

Before this patch, using `ActiveRecord::Base.primary_key_prefix_type`
with `:table_name_with_underscore` would change the `SchemaMigration` model
to have a primary key. This resulted in broken queries for PG because it tried
to return the inserted PK (which does not exist).

Closes #15051.
Closes #15419.

/cc @zuhao

This adds a regression test for #14411, which was fixed by #15503.

Closes #14411
Closes #14595

The migration numbers were normalized different ways. This left
the task output in an inconsistent state.

Closes #15538.

Whiny nils is no longer a thing, so we no longer need this optimization

BC era year is (astronomical year + 1) and starts from 1 BC.

`InverseHasManyTests#test_parent_instance_should_be_shared_within_create_block_of_new_child`
was mistakenly the same as
`InverseHasManyTests#test_parent_instance_should_be_shared_within_build_block_of_new_child`.

Follow up to #15438 and #15502.

/cc @sgrif

For any type that is represented as a string and then type cast, we do
not need separate regular expressions for the various types. No function
will match this regex. User defined types *should* match this, so that
the type object can decide what to do with the value.

`@raw_attributes` should not contain the type-cast, mutable version of
the value.

Reflection has a `belongs_to?` method. Instead of checking for
`macro == :belongs_to` throughout the source reuse existing
method.

I also bumped `foreign_key_present?` method onto on line because
the `belongs_to?` makes it shorter than other longer lines in
the same class.

Revert "test pg, remove unused column assignments. Follow up to 254cdf47"

Related to #15492

This reverts commit 254cdf47.
This reverts commit 4bcf9029.

There are public methods that assume `Column#default` is type casted.
The return value of `Column#default` is publicly relevant and should not change.

/cc @sgrif

than existing column.

Fixes #15480.

The contract of `_field_changed?` assumes that the old value is always
type cast. That is not the case for the value in `Column#default` as
things are today. It appears there are other public methods that
assume that `Column#default` is type cast, as well. The reason for this
change originally was because the value gets put into `@raw_attributes`
in initialize. This reverts to the old behavior on `Column`, and updates
`initialize` to make sure that the values are in the right format.

The types know more about what is going on than the dirty module. Let's
ask them!

That data is internal to Active Record. What we care about is that
new records have the right default value.

/cc @sgrif

Fix polymorphic to check for `options[:polymorphic]` instead of
`options.key? :polymorphic` and then reuse the method `polymorphic?`
method instead of constantly checking the same `options[:polymorphic]`.