This is leftover from when `render nothing: true` rendered blank string.

Use the methods rack provides so we don't have to worry about the exact
header key.

Rails 4.x and earlier didn't support `Mime::Type[:FOO]`, so libraries
that support multiple Rails versions would've had to feature-detect
whether to use `Mime::Type[:FOO]` or `Mime::FOO`.

`Mime[:foo]` has been around for ages to look up registered MIME types
by symbol / extension, though, so libraries and plugins can safely
switch to that without breaking backward- or forward-compatibility.

Note: `Mime::ALL` isn't a real MIME type and isn't registered for lookup
by type or extension, so it's not available as `Mime[:all]`. We use it
internally as a wildcard for `respond_to` negotiation. If you use this
internal constant, continue to reference it with `Mime::ALL`.

Ref. efc6dd55

Just a slight refactor that delegates file sending to the response
object.  This gives us the advantage that if a webserver (in the future)
provides a response object that knows how to do accelerated file
serving, it can implement this method.

* Introduce `ActionDispatch::Http::Headers#add` to add a value to
  a multivalued header.
* Move `Response#add_header` upstream: https://github.com/rack/rack/pull/957
* Match upstream `Response#have_header?` -> `#has_header?` name change.

I want to move the header hash to the super request object in order to
consolidate behavior.  We should be switching out buffering strategies
rather than header strategies since things like "mutating headers after
send" is an error in both cases (buffering vs streaming).

We should not mutate headers after the response has been committed.

I'm making this change so that I can construct response objects that
*don't* have the default headers applied.  For example, I would like to
construct a response object from the return value of a controller.

If you need to construct a response object with the default headers,
then please use the alternate constructor:
`ActionDispatch::Response.create`

We should be asking the mime type method for the mime objects rather
than via const lookup

This method is specifically about the content type so lets remove the
parameter.

create a singleton content type that just has nils, so that we don't
have to allocate a content type object all the time.

If someone sets just a charset, but depends on the implicit type from
rendering, this will store a strange content type header that looks like
this:  `; charset=blah`.  This is so that when the content type header
is parsed again, it will return nil for the actual type.

It turns out that the response object never really cares what the mime
type object is, so just use the string.

pull content-type setting to a private method to dry it up.

Instead of storing content type information in an ivar and a header,
lets move to just store the content type info in just the header.

we'll use this method later to lazily parse content type headers.

cookies are always stored as a string in the header hash, so it will not
need to be joined.

eventually we will refactor this to not know about header hashes

Rack::Response::Helpers implements this method, so we can safely remove
it

Controllers should always have a request and response when responding.
Since we make this The Rule(tm), then controllers don't need to be
somewhere in limbo between "asking a response object for a rack
response" or "I, myself contain a rack response".  This duality leads to
conditionals spread through the codebase that we can delete:

  * https://github.com/rails/rails/blob/85a78d9358aa728298cd020cdc842b55c16f9549/actionpack/lib/action_controller/metal.rb#L221-L223

People should be free to mutate the header object, but not to set a new
header object.  That header object may be specific to the webserver, and
we need to hide it's internals.

this way we don't have to mutate the instance (as much) when writing a
rack response

spelling fix [ci skip]

example to be consistent [ci skip]

grammatical fix

typo fixes [ci skip]

The bug caused a segfault and you can find more info about it at:
https://bugs.ruby-lang.org/issues/10685.

We did a couple of work arounds, but 2.2.1 rolled out and those aren't
needed anymore.

Here are the reverted commits:

- Revert "Work around for upstream Ruby bug #10685",
  commit 707a4338.

- Revert "Fix segmentation fault in ActionPack tests",
  commit 22e0a22d.

I'm also bumping the Ruby version check to 2.2.1 to prevent future
segfaults.

In f6e293ec we avoided a segfault in the
tests, however I think we should try to avoid the crash, as it may
happen in user code as well.

Here is what I distiled the bug down to:

```ruby
# Rails case - works on 2.0, 2.1; crashes on 2.2
require 'action_dispatch'

ActionDispatch::Response.new(200, "Content-Type" => "text/xml")

# General case - works on 2.0, 2.1; crashes on 2.2
def foo(optional = {}, default_argument: nil)
end

foo('quux' => 'bar')
```

Fixes regression in #18423. Merge default headers for new responses,
but don't merge when creating a response from the last session request.

hat tip @senny ❤

These requires were added only to change deprecation message

This patch uniformizes warning messages. I used the most common style
already present in the code base:

* Capitalize the first word.

* End the message with a full stop.

* "Rails 5" instead of "Rails 5.0".

* Backticks for method names and inline code.

Also, converted a few long strings into the new heredoc convention.

The current style for warning messages without newlines uses
concatenation of string literals with manual trailing spaces
where needed.

Heredocs have better readability, and with `squish` we can still
produce a single line.

This is a similar use case to the one that motivated defining
`strip_heredoc`, heredocs are super clean.

- Also one minor change for documenting url_for method in ActionController::Metal.
[ci skip]