feat: CPU steal-time context + CI real-run matrix

[BlueBirdBack]

Summary

• CPU steal-time context: new CPU steal (1s sample) informational metric read from /proc/stat (Linux only; skipped on macOS) so host contention on burstable/shared vCPUs is visible in output and result.json. Not scored.

• Regression test: steal% parsing incl. the no-steal-column case. 24/24 tests pass, shellcheck clean.

• Explicit install contract: new --install/INSTALL=1 flag (default stays on — primary use case is a fresh VPS missing deps) and the pre-install line now names the mechanism: via your system package manager with sudo: <tools> instead of a vague 'attempting install'.

CI upgrade (apply manually — PAT lacks workflow scope)

The workflow change adding a macOS+ubuntu matrix and a real end-to-end run (bash abs.sh --quick -n --no-network, asserting FULL) could not be pushed. Apply the attached diff to .github/workflows/test.yml:

 .github/workflows/test.yml	2026-10-11 03:24:10
+++ /tmp/test-new.yml	2026-10-11 03:24:19
@@ -9,13 +9,47 @@
 
 jobs:
   regression:
-    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-latest, macos-latest]
+    runs-on: ${{ matrix.os }}
     steps:
       - uses: actions/checkout@v5
-      - name: Check shell
+      - name: Check shell (Linux)
+        if: runner.os == 'Linux'
         run: |
           sudo apt-get update
           sudo apt-get install -y shellcheck
           shellcheck abs.sh
+      - name: Check shell (macOS)
+        if: runner.os == 'macOS'
+        run: |
+          brew install shellcheck
+          shellcheck abs.sh
       - name: Run regression suite
         run: python3 -m unittest -v tests/test_abs.py
+
+  end-to-end:
+    # Real benchmark run on real toolchains — the fake-tools suite cannot
+    # catch platform/engine issues (e.g. missing libaio on macOS).
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-latest, macos-latest]
+    runs-on: ${{ matrix.os }}
+    steps:
+      - uses: actions/checkout@v5
+      - name: Install dependencies (Linux)
+        if: runner.os == 'Linux'
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y sysbench fio
+      - name: Install dependencies (macOS)
+        if: runner.os == 'macOS'
+        run: brew install sysbench fio
+      - name: Run ABS quick mode (score must be FULL)
+        run: |
+          set -o pipefail
+          bash abs.sh --quick -n --no-network --lang en --json-file /tmp/abs.json | tee /tmp/abs-out.txt
+          python3 -c 'import json; t=json.load(open("/tmp/abs.json"))["score"]["text"]; assert t.startswith("FULL"), t'

The patch's end-to-end job installs sysbench/fio via apt/brew and asserts score.text starts with FULL — if a future runner filesystem ever lacks O_DIRECT support this is the line to relax.

Devin session: https://app.devin.ai/sessions/c74df4e507504d63aa16cf898640ae58
Requested by: BlueBirdBack