This endpoint is now used in Uppy 2.x.

It doesn't install on GitHub CI.

I believe that rexml was shipping with ruby before, but now it's not
anymore, and aws-sdk-core requires an XML library. So we add it to
development dependencies.

Make it so uploads that don't exist return a 404 error instead of raising an Aws::S3::Errors::NoSuchUpload error. (#21)

There are no changes in this release, we just mark the library as
stable.

Uppy makes requests to the OPTIONS route before the upload, so we
add a handler that just returns a successful response.

Fixes #12

Even though the AWS S3 documentation states that we should do it,
neither Uppy Companion nor tusd are doing this and nobody complained, so
apparently there is no need. If it was really that important, I would
expect aws-sdk-s3 to have a method for "safe aborting".

Since it's good to take every opportunity to make code simpler, we
remove the code that retries aborting the multipart upload.

This was changed in Uppy 1.0

Makes for a nicer `rake routes` command output in Rails apps.

Unlike some other web frameworks like Roda, Rails adds a trailing slash
to PATH_INFO when calling mounted apps on root requests. This means that
for `POST /s3/multipart`, the Uppy::S3Multipart::App will receive

  SCRIPT_NAME: "/s3/multipart"
  PATH_INFO: "/"

instead of

  SCRIPT_NAME: "/s3/multipart"
  PATH_INFO: ""

So we update the app to handle that trailing slash.

:create_multipart_upload and :object_url are probably the only options
that will be used, so it makes sense to show them in the examples.

This allows defining other /s3/* routes without potential clashes when
Uppy::S3Multipart::App isn't mounted last. One example of such route is
"/s3/params", which is called by the AWS S3 Uppy plugin.

We still keep backwards compatibility for users that currently have it
mounted on "/s3".