Fixed a bug in the header parsing code triggered when a continuation line

is a single space: the entire header field above it was ignored and the
message terminated. This would normally only occur when cut-n-pasting from
an xterm, but it revealed a bunch of other problems with the way we were
looking for the request-line and header fields.  I fixed it by rewriting
the routines so that they would be more natural and not discard the
error conditions, and introduced a new routine to buff.* for looking
at the top of the input stream without changing it.

The good news is that we now handle input lines robustly and do not
puke when an extra CRLF is placed at the end of a POST request.
The bad news is that I have found several places where we should be
responding with 400 or 414 (Request-URI Too Long) messages when our
input buffer fills-up, but we have no apparent mechanism for die'ing
inside http_protocol.c.  For now I am just letting the input be truncated,
which is good enough for robust handling, but I would prefer to send an
error response so that the webmaster can know about the invalid
buffer-blowing attempts.


git-svn-id: https://svn.apache.org/repos/asf/httpd/httpd/branches/1.3.x@76971 13f79535-47bb-0310-9956-ffa450edef68