Add golden-file regression test for sysstat-post-process

[atheurer]

Background

Post-processor scripts convert collected system data to CDM metrics via
`CDMMetrics.log_sample()`, writing `metric-data-*.{csv,json}.xz` output.
Without a regression test, changes to the post-processor (optimizations,
new metrics, refactors) can silently produce wrong output.

This issue tracks adding a golden-file regression test for sysstat-post-process.

What to build

```
test/
data/ # committed sample collector output (real, anonymized)
sar/
sar-.xz # sar output files # tool-specific input files
golden/ # committed known-good post-processor output
metric-data-.{csv,json}.xz # generated once from trusted code, committed
run-test.sh # run post-processor on data/, compare to golden/
make-golden.sh # (re)generate golden/ from data/ — run manually, never by CI
```

Test data

Collect a small but realistic sample by running the tool against a real system
for a short interval (30–60 seconds is sufficient). Truncate to a small time
range if needed. The data should:

• Exercise the normal code paths (non-trivial, non-degenerate input)
• Be small enough to commit to git (target < 2 MB uncompressed per input file)
• Not contain sensitive hostnames or credentials

run-test.sh

```bash
#!/bin/bash
set -e
SCRIPT_DIR=$(cd "$(dirname "$0")" && pwd)
TMPDIR=$(mktemp -d)
trap "rm -rf $TMPDIR" EXIT

cp -r "$SCRIPT_DIR/data/." "$TMPDIR/"
(cd "$TMPDIR" && TOOLBOX_HOME=... python3 "$SCRIPT_DIR/../sysstat-post-process.py")

python3 -m toolbox.test_cdm_output "$TMPDIR" "$SCRIPT_DIR/golden/"
```

Comparison semantics (handled by toolbox.test_cdm_output)

• Same metric type sets: every `(desc.type, names)` tuple matches, compared
  by name not by idx (idx is an internal implementation detail).
• Same non-zero values: every non-zero CSV row in golden appears in the
  candidate at the same `(metric_name, end_timestamp, value)`.
• Tolerate leading zeros: zero-value rows before a metric's first active
  period may be absent from the candidate; this is acceptable.

make-golden.sh

Run manually when output intentionally changes (new metric, renamed field).
Commits the updated golden. CI never regenerates golden — only compares.

Acceptance criteria

• `test/data/` committed with realistic sample input
• `test/golden/` committed with output from current (trusted) post-processor
• `test/run-test.sh` exits 0 when output matches golden, non-zero on mismatch
• `test/make-golden.sh` regenerates golden from data
• CI runs `test/run-test.sh` on every PR (add to existing workflow or new one)

Dependencies

Requires toolbox#127 (shared `test_cdm_output.py` comparison utility).