Previously if HTTP/2 traffic is appended to HTTP Upgrade response header
(thus they are in the same buffer), the trailing HTTP/2 traffic is not
processed and lost.  The appended data is most likely SETTINGS frame.
If it is lost, nghttp2 library complains server does not obey the HTTP/2
protocol and issues GOAWAY frame and curl eventually drops connection.
This commit fixes this problem and now trailing data is processed.

It isn't used anywhere!

Reported-by: Carlo Wood

Fix detection of the AsynchDNS feature which not just depends on
pthreads support, but also on whether USE_POSIX_THREADS is set or not.
Caught by test 1014.

This patch adds a new ENABLE_THREADED_RESOLVER option (corresponding to
--enable-threaded-resolver of autotools) which also needs a check for
HAVE_PTHREAD_H.

For symmetry with autotools, CURL_USE_ARES is renamed to ENABLE_ARES
(--enable-ares). Checks that test for the availability actually use
USE_ARES instead as that is the result of whether a-res is available or
not (in practice this does not matter as CARES is marked as required
package, but nevertheless it is better to write the intent).

Signed-off-by: Peter Wu <peter@lekensteyn.nl>

vtls.c:185:46: warning: unused parameter 'data'

In preparation for moving the NTLM message code into the SASL module,
and separating the native code from the SSPI code, added functions that
simply call the functions in curl_ntlm_msg.c.

In preparation for moving the NTLM message code into the SASL module
use the SASL functions in the HTTP code instead.

... so that non-configure builds get the correct functions too based on
the libssh2 version used.

... if the libssh2 version is new enough.

Bug: http://curl.haxx.se/bug/view.cgi?id=1448

USE_NTLM would only be defined if: HTTP support was enabled, NTLM and
cryptography weren't disabled, and either a supporting cryptography
library or Windows SSPI was being compiled against.

This means it was not possible to build libcurl without HTTP support
and use NTLM for other protocols such as IMAP, POP3 and SMTP. Rather
than introduce a new SASL pre-processor definition, removed the HTTP
prerequisite just like USE_SPNEGO and USE_KRB5.

Note: Winbind support still needs to be dependent on CURL_DISABLE_HTTP
as it is only available to HTTP at present.

This bug dates back to August 2011 when I started to add support for
NTLM to SMTP.

Due to commit 40ee1ba0 the free in Curl_ntlm_decode_type2_target() is
longer required.

Fixed a couple of memory leaks as a result of moving code that used to
populate allocuserpwd and relied on it's clean up.

Following the addition of SSPI support for HTTP digest, synchronised
elements of the email digest code with that of the new HTTP code.

Post tidy up to ensure commonality of code style and variable names.

Reworked the input token (challenge message) storage as what is passed
to the buf and desc in the response generation are typically blobs of
data rather than strings, so this is more in keeping with other areas
of the SSPI code, such as the NTLM message functions.

Added void reference to unused 'data' parameter back to fix compilation
warning.

Some versions of Microsoft's sspi.h don't define this.

Introduced in commit 7e6d51a7 these functions and definitions are only
required by the internal challenge-response functions now.

This temporarily breaks HTTP digest authentication in SSPI based builds,
causing CURLE_NOT_BUILT_IN to be returned. A follow up commit will
resume normal operation.

Added forward declaration of digestdata to overcome the following
compilation warning:

warning: 'struct digestdata' declared inside parameter list

Additionally made the ntlmdata forward declaration dependent on
USE_NTLM similar to how digestdata and kerberosdata are.

Broken as part of the rework, in commit 7e6d51a7, to assist with the
addition of HTTP digest via Windows SSPI.

error: invalid operands to binary
warning: pointer targets in assignment differ in signedness

Bug: http://curl.haxx.se/mail/lib-2014-11/0035.html
Reported-by: Jan Ehrhardt

...as Curl_memdup() duplicates an area of fix size memory, that may be
binary, and not a null terminated string.

conversion from 'curl_off_t' to 'size_t', possible loss of data