multi: fix slow write/upload performance on Windows

Reset FD_WRITE by sending zero bytes which is permissible
and will be treated by implementations as successful.

Bug: curl#6146
Closes curl#6245

Author: mback2k

lib/multi.c

@@ -1177,22 +1177,23 @@ static CURLMcode Curl_multi_wait(struct Curl_multi *multi,
         long mask = 0;
 #endif
         if(bitmap & GETSOCK_READSOCK(i)) {
+          s = sockbunch[i];
 #ifdef USE_WINSOCK
           mask |= FD_READ|FD_ACCEPT|FD_CLOSE;
 #endif
-          ufds[nfds].fd = sockbunch[i];
+          ufds[nfds].fd = s;
           ufds[nfds].events = POLLIN;
           ++nfds;
-          s = sockbunch[i];
         }
         if(bitmap & GETSOCK_WRITESOCK(i)) {
+          s = sockbunch[i];
 #ifdef USE_WINSOCK
           mask |= FD_WRITE|FD_CONNECT|FD_CLOSE;
+          send(s, NULL, 0, 0); /* reset FD_WRITE */
 #endif
-          ufds[nfds].fd = sockbunch[i];
+          ufds[nfds].fd = s;
           ufds[nfds].events = POLLOUT;
           ++nfds;
-          s = sockbunch[i];
         }
         if(s == CURL_SOCKET_BAD) {
           break;
@@ -1218,8 +1219,10 @@ static CURLMcode Curl_multi_wait(struct Curl_multi *multi,
       mask |= FD_READ|FD_ACCEPT|FD_CLOSE;
     if(extra_fds[i].events & CURL_WAIT_POLLPRI)
       mask |= FD_OOB;
-    if(extra_fds[i].events & CURL_WAIT_POLLOUT)
+    if(extra_fds[i].events & CURL_WAIT_POLLOUT) {
       mask |= FD_WRITE|FD_CONNECT|FD_CLOSE;
+      send(extra_fds[i].fd, NULL, 0, 0); /* reset FD_WRITE */
+    }
     if(WSAEventSelect(extra_fds[i].fd, multi->wsa_event, mask) != 0) {
       if(ufds_malloc)
         free(ufds);