Only test ones we know will succeed (#1628)

A bit of a hack to only run the network tests we expect to succeed. This
at least ensures we don't get any worse, even if it doesn't directly
allow us to track if we're getting better.

See also actions/runner#2347

.github/workflows/network-test.yaml

@@ -29,7 +29,8 @@ jobs:
         # Currently this is just a label and does not have any functional impact.
         peers:          [3]
         scaling_factor: [10, 50]
-        netem_loss:     [0, 1, 2, 3, 4, 5, 10, 20]
+        # Note: We only put here the configuration values we _expected to pass_.
+        netem_loss:     [0, 1, 2, 3]
     name: "Peers: ${{ matrix.peers }}, scaling: ${{ matrix.scaling_factor }}, loss: ${{ matrix.netem_loss }}"
     steps:
     - uses: actions/checkout@v4
@@ -61,6 +62,8 @@ jobs:
 
     - name: Setup containers for network testing
       run: |
+        set -exo pipefail
+
         cd demo
         ./prepare-devnet.sh
         docker compose up -d cardano-node
@@ -73,9 +76,10 @@ jobs:
           --node-socket devnet/node.socket \
           --cardano-signing-key devnet/credentials/faucet.sk)
 
-        echo $HYDRA_SCRIPTS_TX_ID >> .env
+        echo "HYDRA_SCRIPTS_TX_ID=$HYDRA_SCRIPTS_TX_ID" > .env
 
-        nix run .#cardano-cli query protocol-parameters \
+        nix run .#cardano-cli -- query protocol-parameters \
+          --testnet-magic 42 \
           --socket-path devnet/node.socket  \
           --out-file /dev/stdout \
           | jq ".txFeeFixed = 0 | .txFeePerByte = 0 | .executionUnitPrices.priceMemory = 0 | .executionUnitPrices.priceSteps = 0" \
@@ -103,14 +107,6 @@ jobs:
         limit-access-to-actor: true
 
     - name: Run pumba and the benchmarks
-      # Note: We're going to allow everything to fail. In the job on GitHub,
-      # we will be able to see which ones _did_, in fact, fail. Originally,
-      # we were keeping track of our expectations with 'include' and
-      # 'exclude' directives here, but I think it's best to leave those out,
-      # as some of the tests (say 5%) fail, and overall the conditions of
-      # failure depend on the scaling factor, the peers, etc, and it becomes
-      # too complicated to track here.
-      continue-on-error: true
       run: |
         # Extract inputs with defaults for non-workflow_dispatch events
         percent="${{ matrix.netem_loss }}"