Simplify ucxx::RequestTagMulti final status logic (#144)

The current way to handle the final status of `ucxx::RequestTagMulti` involves looping all requests, this change simplifies that by simply caching the status of the first request error. Additionally this prevents checking the status of the final request if it hasn't been set yet, which would then return `UCS_INPROGRESS` instead.

Authors:
  - Peter Andreas Entschev (https://github.com/pentschev)

Approvers:
  - Lawrence Mitchell (https://github.com/wence-)

URL: #144

cpp/include/ucxx/request_tag_multi.h

@@ -44,6 +44,8 @@ class RequestTagMulti : public Request {
   std::mutex
     _completedRequestsMutex{};   ///< Mutex to control access to completed requests container
   size_t _completedRequests{0};  ///< Count requests that already completed
+  ucs_status_t _finalStatus{
+    UCS_OK};  ///< Shortcut to the final status, a.k.a. the first error to occur
 
  public:
   std::vector<BufferRequestPtr> _bufferRequests{};  ///< Container of all requests posted

cpp/src/request_tag_multi.cpp

@@ -154,28 +154,27 @@ void RequestTagMulti::markCompleted(ucs_status_t status, RequestCallbackUserData
   ucxx_trace_req("RequestTagMulti::markCompleted request: %p, tag: %lx", this, _tag);
   std::lock_guard<std::mutex> lock(_completedRequestsMutex);
 
+  if (_finalStatus == UCS_OK && status != UCS_OK) _finalStatus = status;
+
   if (++_completedRequests == _totalFrames) {
-    auto s = UCS_OK;
-
-    // Get the first non-UCS_OK status and set that as complete status
-    for (const auto& br : _bufferRequests) {
-      if (br->request) {
-        s = br->request->getStatus();
-        if (s != UCS_OK) break;
-      }
-    }
-    // Check the status of the current (and final) message as it may have completed before
-    // `_bufferRequests->request` was populated.
-    if (s == UCS_OK) s = status;
+    setStatus(_finalStatus);
 
-    setStatus(s);
+    ucxx_trace_req(
+      "RequestTagMulti::markCompleted request: %p, tag: %lx, completed: %lu/%lu, final status: %d "
+      "(%s)",
+      this,
+      _tag,
+      _completedRequests,
+      _totalFrames,
+      _finalStatus,
+      ucs_status_string(_finalStatus));
+  } else {
+    ucxx_trace_req("RequestTagMulti::markCompleted request: %p, tag: %lx, completed: %lu/%lu",
+                   this,
+                   _tag,
+                   _completedRequests,
+                   _totalFrames);
   }
-
-  ucxx_trace_req("RequestTagMulti::markCompleted request: %p, tag: %lx, completed: %lu/%lu",
-                 this,
-                 _tag,
-                 _completedRequests,
-                 _totalFrames);
 }
 
 void RequestTagMulti::recvHeader()