ap_process_http_async_connection (used by the Event MPM):

fix CLOSE_WAITs/leaked connections.  The logic to deal with ap_read_request
failures got lost when merging in the HTTP pipelining fix.  If ap_read_request
fails, the connection state should get set to CONN_STATE_LINGER so the MPM will
invoke lingering close.  Test case: client sends a Connection: keepalive header
then closes the connection before the keepalive timeout pops.

Also add a comment to make the pipelining flow more obvious.


git-svn-id: https://svn.apache.org/repos/asf/httpd/httpd/trunk@106662 13f79535-47bb-0310-9956-ffa450edef68