Formatting from header acts like Versioning from header

[ericproulx]

This PR changes the how the content_type is negotiated in Grape::Middleware::Formatter when looking in the ACCEPT header. Since #2389, Grape::Middleware::Versioner::Header is using Rack::Utils.best_q_match to find it but not Grape::Middleware::Formatter. It seems totally legit that both should act the same way.

This PR also optimize the negotiation regarding the extension found in a path lie /api/v2/test.json. Originally, it was using split without limit the number of occurrences that could be found and since its always looking at the end of string , it felts natural to just find the first dot starting from the end of the string.

Finally, Grape::Middleware::Formatter won't approximate the content-type. This test was expecting an approximation when using a vendor/type form

it 'is set to closest generic for custom vendored/versioned without registered type' do
   _, headers, = subject.call(Rack::PATH_INFO => '/info', Grape::Http::Headers::HTTP_ACCEPT => 'application/vnd.test+json')
   expect(headers[Rack::CONTENT_TYPE]).to eq('application/json')
end

IMO, it didn't make sense since user should use versioning through Grape::Middleware::Versioner in their API. It was made for this.

[dblock]

Since there's definite change in behavior (even if that behavior was not desired, invalid, etc.) this needs a clear UPGRADING.md. Go through the tests being removed and try to use those as examples?

[ericproulx]

@dblock done