DPL Analysis: look up the CCDB output buffer only when fetching

populateCacheWith created the pmr vector for the object, which matches
the output route and looks up its transport, for every FetchOp. On a
cache hit the vector is never used, and the analysis CCDB fetcher issues
one FetchOp per column for every row of the source table. Create it
only where an object is actually requested from the server.

Author: ktf

Framework/CCDBSupport/src/CCDBFetcherHelper.cxx

@@ -179,7 +179,6 @@ auto CCDBFetcherHelper::populateCacheWith(std::shared_ptr<CCDBFetcherHelper> con
     O2_SIGNPOST_EVENT_EMIT(ccdb, sid, "populateCacheWith", "Fetching object for route %{public}s", DataSpecUtils::describe(op.spec).data());
     auto concrete = DataSpecUtils::asConcreteDataMatcher(op.spec);
     Output output{concrete.origin, concrete.description, concrete.subSpec};
-    auto&& v = allocator.makeVector<char>(output);
     std::map<std::string, std::string> metadata;
     std::map<std::string, std::string> headers;
     std::string path = op.url;
@@ -232,6 +231,7 @@ auto CCDBFetcherHelper::populateCacheWith(std::shared_ptr<CCDBFetcherHelper> con
 
     const auto& api = helper->getAPI(path);
     if (checkValidity && (!api.isSnapshotMode() || etag.empty())) { // in the snapshot mode the object needs to be fetched only once
+      auto&& v = allocator.makeVector<char>(output);
       LOGP(detail, "Loading {} for timestamp {}", path, timestampToUse);
       api.loadFileToMemory(v, path, metadata, timestampToUse, &headers, etag, helper->createdNotAfter, helper->createdNotBefore);
       if ((headers.count("Error") != 0) || (etag.empty() && v.empty())) {