When we receive a "bad header" we must sure not to write down the data part
as well, as then we write the same data twice.

Junk data could get inserted when saving/getting HTTP headers, as discovered
by Craig Davison. Now we deal with the 'nread' variable correctly between
each header line.

Move the URL concat code to Curl_follow(), and added a proto for that
function. For Location: following.

as Ralph Mitchell pointed out, the Location: following code needs some
basic ./ and ../ strip-off understanding, and this change introduces with.
test cases 49 - 52 test this.

spell fix comment

a follow-up fix to the previous fix for HTTP servers that don't reply *any*
headers at all

better deal with HTTP(S) servers that respond with no headers at all, test
case 306 added to verify that we do right

remove redundant comment

updated source code boilerplate/header

James Gallagher's Content-Encoding work

remove the data and conn fields from the Curl_transfer_keeper struct, they
weren't used anyway and mostly caused confusion

Markus F.X.J. Oberhumer's CURLOPT_NOSIGNAL patch

Make SessionHandle keep record if it is used with the multi interface or
the easy interface, it CANNOT be used by a mixture.

fixed the Curl_cookie_add() invoke to not assume a space after the colon

CURLOPT_BUFFERSIZE allows an application to set a prefered buffer size
for receiving data from the network. It is meant as a hint, not as a forced
limit.

Moved the secondarysocket cleanup. The 'conn' struct may have been cleared
already at that other place.

Gautam Mani found a socket descriptor leak that happened when FTP transfers
failed and you reinvoked curl_easy_perform().

Adjusted to make curl_multi_perform() work properly even when
curl_multi_fdset() is not used.

Better support for being used with the multi interface without the *fd_set()
and proper select()ing have been made.

made the DEBUGFUNCTION get called properly on a few more places, especially
for DATA_IN and DATA_OUT.

use and set the fd_set pointers instead of the actual values, as then we
work properly with the multi interface when the user has provided the
fd_sets!

Hanno Kranzhoff noticed we didn't properly reset the download/upload counters
before transfers, when doing multiple ones on the same handle.

Dirk Manske's fix for HTTP response code 301, that now behaves more like
browsers do. Which thus *breaks* the RFC 2616...

Dirk Manske brought the patch that introduces two new CURLINFO_* values:
CURLINFO_REDIRECT_TIME and CURLINFO_REDIRECT_COUNT.

Jonatan Lander fixed the "Disables POST, goes with GET" output to be more
connected with reality! ;-)