Rework gitlab reply such that order of ref and mr build do not interfere, also...
This is a rework of !178 (merged) that fixes the fact that the ThroughputProfileHandler.py can't rely on whether the throughput test is first run for lhcb-master-ref or lhcb-master-mr.
Additionally, gathering results is now done only via couchDB as the results aren't available instantly via LHCbPR2.
This meant that even if a slot was running second, it could potentially not find the results for the other slot via the LHCbPR2 API.