Run one benchmark at a time (#1585)

Before the change, benchmarks ran in parallel. We are observing very
high variance no the benchmarks, part of the problem could be resource
contention for network bandwidth across benchmarks. This change will
give the entire runner's worth of resources to each benchmark
individually, and hopefully reduce variance at the cost of using more CI
runner time. Will observe for 1 week and evaluate.

.github/workflows/performance_metrics_cron.yml

@@ -90,7 +90,7 @@ jobs:
         shell: bash
         run: |
           set -euo pipefail
-          cd misc/test && go test . --timeout 4h -count=1 -json -short -parallel 40 --tags=Performance 2>&1 | tee /tmp/gotest.log | gotestfmt
+          cd misc/test && go test . --timeout 4h -count=1 -json -short -parallel 1 --tags=Performance 2>&1 | tee /tmp/gotest.log | gotestfmt
       - name: Upload traces GHA artifact
         uses: actions/upload-artifact@v3
         with: