Only run test case for http as the socket manipulation done by the test case
requires much more manipulation to work for https. However, that this test case
started failing constantly instead of sporadically is proof that this change
is a good thing.

If a message is received with both a Transfer-Encoding and a
Content-Length header field, it might indicate an attempt to
perform request smuggling or response splitting and must be
handled as an error in default mode (not relaxed mode).

Bug report: https://bugs.erlang.org/browse/ERL-407

Guard test for error case was not updated to handle httpc stream concept
properly.

Probably since 6153ba7 (OTP R13B04) the httpc timeout setting does not
work for redirects (when autoredirect is true). With this patch a new
timer is started for the new (redirected) requests. This means that a
simple redirected request could return after 2*timeout milliseconds.

This is the first part to fix https://bugs.erlang.org/browse/ERL-420

Conflicts:
	OTP_VERSION
	erts/vsn.mk
	lib/crypto/c_src/crypto.c
	lib/crypto/src/crypto.erl
	lib/ssh/src/ssh.erl

ERL-316, as part of 19.3, adds the port number to the Host header
upon automatic redirection. The port number is included even if it
is a well-known port (80, 443). This is different from the
behaviour of most HTTP clients, as well as httpc's own for new
requests.

The added port number can lead to problems such as this one, where
the request signature assumes the client will not send the :443
suffix on redirection to an https URL:
https://github.com/nerves-project/nerves/issues/96

I was unable to add a test case, since that would require a server
on a well-known port, but I manually verified that the GitHub/S3
signing issue was indeed resolved with this patch.