pm-bench stats <name> — one-shot summary stats for any log

[protosphinx]

Stacked on top of #12 (synthetic-200). Merge order: #2 → #3 → #4 → #5 → #6 → #7 → #8 → #9 → #10 → #11 → #12 → this.

Summary

• New CLI: pm-bench stats <name-or-path> prints n_cases / n_events / n_activities / time span / mean+median case length / top-N activities / top-N transitions.
• Pure CPython; works on synthetic-toy and any CSV path. Dispatched through the same _load_events so the CSV-ingest path Just Works.

What's new

• pm_bench/stats.py:summarize(events, top_n=10) — one pass over the event iterable; returns LogStats (frozen dataclass).
• CLI pm-bench stats emits JSON, includes earliest/latest ISO timestamps and span_days so you can sanity-check a split before running it.
• README gets a one-liner pointing at the command, right above the CSV ingest snippet.
• 7 new tests including a CLI smoke against synthetic-toy. 116 total, ruff clean.

Smoke

$ pm-bench stats synthetic-toy --top-n 5 | head -10
{
  "n_events": 965,
  "n_cases": 200,
  "n_activities": 9,
  "span_days": 367.21,
  "earliest": "2024-01-01T00:00:00",
  "latest": "2025-01-02T05:00:00",
  "mean_case_length": 4.825,
  ...
}

Why

• Anyone trying pm-bench on their own CSV currently has no quick "is my log loaded right?" check. stats answers that in one command before they wire up a model.
• It's also useful for the leaderboard narrative: every entry can include the dataset's stats inline, without re-running PM4Py.

Test plan

• pytest -q — 116 passed (was 109 on PR synthetic-toy → 200 cases: outcome leaderboard row lands; all 5 boards real #12)
• ruff check pm_bench tests — clean
• CLI smoke against synthetic-toy returns expected counts (200 cases, 965 events, 9 activities)

[protosphinx]

Merged into main as part of the audit-cleanup stack (commit 9c00b47). The full content of this PR is now on main.