Rubocop / code climate don't like single quotes and prefer doubles.

Puma is the default webserver of Rails. Because of this it doesn't make
sense to run tests in Webkit if the default server is Puma.

Here I've refactored the webserver to be it's own standalone module so
it can be shared between Rails' selenium default driver and Capybara's
defaut drivers.

This change adds support, tests, and documentation for the screenshot
helper.

If taking screenshots is supported by the driver (for example Rack Test
doesn't support screenshots) then a screenshot will be taken if the test
fails.

This removes the useless Rack Test Driver that Rails was providing and
moves to a shim like approach for default adapters.

If someone wants to use one of the default Capybara Drivers then we will
initialize a new `CapybaraDriver` that simply sets the default driver.

Rails though is much more opinionated than Capybara and to make system
testing a "works out of the box" framework in Rails we have the
`RailsSeleniumDriver`. This driver sets defaults that Rails deems
important for selenium testing. The purpose of this is to simply add a
test and it just works.

* Document Rails::SystemTestCase
* Document setting drivers with the configration options
* Document using the getter/setter for driver adapters
* Document the CapybaraRackTestDriver and defaults
* Document the CapybaraSeleniumDriver and defaults
* Document custom assertions provided by System Testing
* Document custom form helpers provided by System Testing

* Adds test case test
* Adds driver adapter test
* Adds tests for capybara seleium driver (testing the settings not
actually opening the browser to test capybara w/ selenium because that
would so so so slow)
* Adds tests for rack test driver
* Adds tests for generators

This will clean up the railtie quite a bit, rather than passing a set of
hash keys call the new class directly like we do with ActiveJob.

Only call driver once when tests start rather than in every single test
setup. This is more performant, and the other way was creating
unnecessary calls.

There's no real benefit to the using the `Base` class here because
`SystemTestCase` is already a very small class.

Rails itself is not a Rails application so instead of including
`capybara/rails` we should use the code in there to set up the test. The
only reason capybara needs to include capybara/rails in the first place
is because Rails didn't yet support it.

This makes it easier to ask the system test what driver adapter it is
currently using, and makes it easier to change that setting when
necessary.

Integration tests already handle all the fancy url mapping we need to do
so inherting from that allows us to not need to reinvent the wheel in
terms of loading up the route handling required to use `visit
users_path` over `visit /users`.

Adds assertions that are not part of Capybara but may be useful to Rails
users writing system tests.

This allows any application to change the driver adapter based on the
config settings in the test env.

These FormHelpers are selectors that aren't a capybara default but are
considered useful for Rails applications.

This is not yet configurable but is the minimum required to make
Capybara work with the Selenium driver. A lot of this will change as the
tests get fleshed out and the initialization requirements will eventually
be configurable via the application.

Capybara defaults to Rack Test for it's driver and works out of the box
but this adds the headers and allows for future configurable adapters
for system testing.

This skelton is the bare minimum to get system tests to actually run in
an application. This of course doesn't yet actually run a test but it is
enough for `bin/rails test:system` to attempt to run files in
`test/system` that inherit from `Rails::SystemTestCase`.

* Generates system test requirements with new Rails app
* Includes required default gems in Gemfile for Rails app
* Generates a single system test case
* Generates a system test case with scaffold

Some improvements to the API site's sidebar

Delegate to `scope` rather than `merge!` for collection proxy

Since Active Support is monkey patching a lot of core classes, let's
rather document these changes under a new section so they are still
documented but not encumbering the sidebar.

We can safely remove the rescuing of the `LoadError` since as of
cd7cc525, it's not possible to generate the API from an application.

[ci skip]

[Kasper Timm Hansen & Robin Dupret]

Update Guides to use macOS instead of Mac OS X [ci skip]

Remove the top-level `HashWithIndifferentAccess` contant

Do not run `git init` in dummy application

Improve `create_table force: true`

Add assertion to polymorphic_routes_test.rb

The assertion will ensure that the behavior doesn't regress.
  assert_equal "/projects", polymorphic_path("projects")

Remove FIXME related to polymorphic_url behavior.
polymorphic_url with Symbol or String works equally.
Example:

  default_url_options[:host] = "example.com"

  polymorphic_url(:projects)  # => "http://example.com/projects"
  polymorphic_url("projects") # => "http://example.com/projects"

Related to 37d4415a

This constant was kept for the sake of backward compatibility; it
is still available under `ActiveSupport::HashWithIndifferentAccess`.

Furthermore, since Ruby 2.5 (https://bugs.ruby-lang.org/issues/11547)
won't support top level constant lookup, people would have to update
their code anyway.

Add missing module namespacing wrapper

Unfreeze interpolated string because it's unnecessary.

Extra `data_source_exists?(table_name)` is unneeded if
`drop_table(table_name, if_exists: true)` directly.

Remove deprecate passing string to `:if` and `:unless` conditional options [ci skip]

Follow up to #27608