Ensure we submit endpoint information before attempting opt.

This is only needed because of something like async-connect
followed by select (I guess?) and next thing we know we've
sent "N" bytes and still haven't told ipcd about our timings.

We should do this earlier, for sure.

libipc/io.cpp

@@ -101,6 +101,12 @@ void attempt_optimization(int fd, bool send) {
   if (get_byte_counter(i, send) == TRANS_THRESHOLD) {
     ipclog("Completed partial operation to sync at THRESHOLD for fd=%d!\n", fd);
 
+    // If we're here, we definitely should have already submitted info.
+    // XXX: This happens if first IO operation causes us to cross our
+    // threshold.  This fixes it for now, but should be done earlier.
+    submit_info_if_needed(fd);
+
+
     // TODO: Async!
     endpoint remote = EP_INVALID;
     size_t attempts = 0;