Use express response methods

[mike-marcacci]

We should not bypass the express convention of using res.send here. Doing so prevents other middleware from working as intended.

This includes graphql-upload, which is currently unable to force the response to wait for the request to be fully consumed, and causes severe bugs, including locking a browser out of subsequent requests to the server.

Please see this issue and the accompanying demonstration repo put together with heroic effort by @juona.

[jaydenseric]

Does this also need to be fixed in the other middleware packages such as apollo-server-koa?

[mike-marcacci]

The koa middle correctly follows the koa convention of using ctx.body = graphqlResponse to set the response body, so the behavior is correct when someone brings their own instance of graphql-upload.

However, all of the upload middleware that are build in to the various apollo-server- packages appear to lack any sort of waiting, so this bug will exist for anyone using uploads but not directly using the middleware exported by graphql-upload

I created issue #2843 for this separate, larger problem.

[mike-marcacci]

Well, it appears that this package is also tested with connect, and not just express. I'll just leave this here for the Apollo team to decide on, as I don't use any of this anymore.

[abernix]

LGTM / Thanks, @mike-marcacci. I certainly agree that using send here is more appropriate.

I've added a commit which falls back to calling end when send isn't available, which should satisfy connect's desires (and if tests are any barometer, the additional commit has resolved the failures).

As you've noted in your other PR, there's certainly room for additional improvement here, and while this code has gone relatively unchanged for almost 3 years, this change should still be an improvement worth getting in ASAP.