Redirect handling not quite right?

[ericphanson]

In #389 I added a throw if we could not set the URL from the returned location, but did not change the behavior otherwise.

But I'm not quite sure that original behavior was right: it seems like we set the URL in the request, but if nothing throws, we just continue on our way. I assumed that we would automatically retry with the new URL, but that was a misunderstanding of @retry. If there is no error, it does not retry. So I don't think setting the URL does much, since we don't use it to make another request.

I'm not really sure though when we will get a redirect from AWS that we want to follow; when you have the wrong region, we get 302s (without a location parameter sent back), but I'm not sure in what other circumstances it happens. On my #396 branch I have seen this code path triggered by

AWSServices.sts(
                             "AssumeRoleWithWebIdentity",
                             Dict(
                                 "RoleArn" => role_arn, # an actual role arn
                                 "RoleSessionName" => "test",
                                 "WebIdentityToken" => "123", # made up
                             );
                             aws_config=AWSConfig(creds=nothing)

but I don't know why-- AWS should be returning an error code about the WebIdentityToken being bad, not a redirect, like it does on master.

Also, I'm not sure why we don't use redirect=true in our HTTP stack, if we do want to follow redirects.

cc @omus