I refactored `AnnotateRoutes.header` for performance, and add `AnnotateRoutes.comment` to make code DRY.

As the change log shows, I froze the constant `HEADER_ROW` in accordance with the normal Ruby style regulation.

I added `AnnotateModels::MAGIC_COMMENT_MATCHER` and `AnnotateRoutes::MAGIC_COMMENT_MATCHER` instad of `AnnotateModels.magic_comment_matcher` and `AnnotateRoutes.magic_comment_matcher`.

I think that they should be constants because the result of Regexp is always fixed.

Before I commited, the method `AnnotateRoutes.where_header_found` returned an array of `real_content` and `header_found_at`.

I fixed the name of method and variable because they did not match.

I renamed the method `real_content_and_header_position` and unified related variable names to `header_position`.

This change converts .all_options into a constant and moves it into Annotate::Constants. It also changes usages of .all_options.

Replace soft-deprecated constant `HashWithIndifferentAccess` to `ActiveSupport::HashWithIndifferentAccess` (#699)

Same problem returned by @janosrusiczki in #430

Spatial columns with activerecord-postgis-adapter (5.2.2) are not displayed nice

```
# name           :string(3)
# active         :boolean          default(FALSE), not null
# geom           :geometry({:srid= geometry, 4326
```

That's come from activerecord-postgis-adapter, with redefined columns as such :

``` 
#<ActiveRecord::ConnectionAdapters::PostGIS::SpatialColumn 
  @sql_type="geometry(Geometry,4326)",
  @geo_type="Geometry",
  @geometric_type=RGeo::Feature::Geometry,
  @srid=4326,
  @limit={:srid=>4326, :type=>"geometry"}
  [...]
>
```

This fix displays them like this :

```
# name           :string(3)
# active         :boolean          default(FALSE), not null
# geometry       :geometry         geometry, 4326
```

Another possibility would have been to display them as below, but it involves a lot of extra-spaces for other columns.

```
# name           :string(3)
# active         :boolean                   default(FALSE), not null
# geometry       :geometry(Geometry, 4326)
```

ActiveAdmin defaults to false.  Add `active_admin` option so people using this gem can more easily enable annotating files in `app/admin`.

Rails 6 has dropped name from ActiveRecord::ConnectionAdapters::ForeignKeyDefinition
for Sqlite3 adapter:

https://github.com/rails/rails/blob/f2df77709f7e536aaf4d6f984ff21a49d44d34c1/activerecord/lib/active_record/connection_adapters/sqlite3_adapter.rb\#L305-L311

This has caused issues with getting foreign key info and blowing up
when trying to annotate models with foreign keys. This commit
adds one additional check for name presence and uses column
attribute from options instead.

Closes #620.

Setting the default to annotate models as it is the main feature of the gem

Changed references to `--additional_file_patterns` to be `--additional-file-patterns` to make it consistent with other flags.

Adds tests for `.include_routes?`, `.include_models?`, `.skip_on_migration?`. Also moves the `TRUE_RE` under the `Annotate::Constants` namespace.

Prior to this change, `Annotate.include_models?` returned the inverse of `Annotate.include_routes?`. This made it so annotating models and routes was not possible to do together.

This PR adds an explicit `--models` flag and also adds it the option to `lib/generators/annotate/templates/auto_annotate_models.rake` with the default being set to `false`. 

Fixes #563 and undoes the bug introduced in #485.

Add better documentation for new option for additional file patterns. Implemented in #633 and #636.

Adds option for additional file patterns (implemented in #633) in the CLI.

Refactored Parser to isolate changes being made to ENV. This way we have an intermediate step where we know the environment variables being set.

This PR adds an option named `additional_file_patterns`.  You can specify custom path patterns (including globs) that the gem will use to annotate.

For example, I used it like this on a project:

```ruby
Annotate.set_defaults(
      'additional_file_patterns' => [
        File.join(Rails.application.root, 'app/lib/forms/%PLURALIZED_MODEL_NAME%/**/*.rb'),
        File.join(Rails.application.root, 'spec/lib/forms/%PLURALIZED_MODEL_NAME%/**/*.rb')
      ],
      ...
)
```

This makes it possible to have files nested under a directory which corresponds to the model.

I believe this fixes #594.

Currently there's no test coverage on CLI. This adds tests for a new class Parser which will replace the command line options currently in bin/annotate.

The technical direction I'm planning to go is to remove ENV variables completely and to have things passed into as arguments. Will be adding deprecation warnings in 3.0 and then deprecate ENV variables completely in 3.1. Deprecation warnings will be non-blocking in 3.0 when an ENV variable is set and an argument isn't passed in, and will becoming blocking in 3.1.

Handle potential negative padding in `#mb_chars_ljust` when having a `col_length` > 16, the default hardcoded value of `bare_type_allowance`.

In AnnotateRoutes, there were many public methods that is not used in other classes or modules.

Before my extension, I made these methods private and sort them in order of appearance.

All tests were passed.

Please trace each commits carefully. I didn't nothing special. The commits consist of coordinating code.

Add --frozen option to abort instead of write annotation.
This is mainly for CI environment to detect missing annotation.
Similar motivation with bundle install --frozen.

works like:
```
$ rake db:migrate frozen=yes

annotate error. spec/models/user_spec.rb needs to be updated, but annotate was run with `--frozen
```

Since Rails full engines run as non-namespaced applications but all their app-related tasks still prefixed with `app:`

- Use Rake::Task#enhance insteaad of defining same tasks again
- Remove hooking db:migrate:change task which doesn't exist
- Fix hooking db:migrate:reset task so that the annotation runs after
  all migration tasks (#548)

This fixes alignment of Japanese, Korean and Chinese fullwidth character comments.

The displayed widths of multi-byte fullwidth characters are generally
twice as large as the ASCII characters, but String#size returns only the number of characters.
So if the column comment contains fullwidth multibyte characters, the alignment is broken.

* refactor and simplify returns for annotate_one_file
* fix rubocop offenses