feat: producer run summary, --check, status.json, and --reconcile

[mspinola]

Producer visibility, in three parts.

1. Per-run console summary

Each --prices run prints a per-symbol date advance and a footer:

ES: 9946 backadj, 9946 unadj  [2026-07-13 -> 2026-07-14]
...
prices update: 41/42 OK · 1 failed · 1,658,000 rows · 88s · newest 2026-07-14
  ✗ GC: <error>

2. cotdata-update --check — read-only status

Per-domain row counts, newest data date, last write, and any entries lagging behind their peers (a partial-run signal). Manifest-only — no network, no Windows/norgatedata — so it runs anywhere the store is visible.

3. status.json — structured new-data signal for downstream tools

Every run writes $COTDATA_STORE/status.json (atomically, beside the data) so tools that trigger on fresh data poll one small file:

{
  "generated_at": "2026-07-15T10:15:24Z",
  "schema_version": 2,
  "newest_data": { "prices": "2026-07-14", "cot_legacy": "2026-07-07", "cot_disagg": "2026-07-07", "cot_tff": "2026-07-07" },
  "domains":     { "prices": { "newest_data": "2026-07-14", "last_write": "...", "entries": 84, "rows": 829096, "lagging": 0 } },
  "last_run":    { "kinds": ["prices"], "ok": ["ES","..."], "symbols_failed": [], "rows": 1658000, "seconds": 88, "at": "..." }
}

Polling contract:

• New data → compare newest_data.<domain> to your last-seen value; it advances only on genuinely new daily data. Prices and each COT report are separate domains, so price- and COT-triggered tools each watch their own key.
• A run happened → generated_at.
• Alerting → last_run (kinds, per-symbol failures).

Notes

• Pure helpers (summarize / format_report / build_status_doc / write_status_file / run_summary) live in status.py; test_status.py covers them. Full suite: 41 passed.
• Email delivery intentionally not built (needs SMTP config/secrets); run_summary returns a string and status.json is structured, so it's a small follow-up.

🤖 Generated with Claude Code

4. cotdata-update --reconcile — manifest hygiene / COT naming consistency

COT parquet files are already named {symbol}_{code} (e.g. RTY_23977A), and get_cot reads that convention. But the manifest carried ghosts from older schemes — bare CFTC codes (001602, 23977A) predating the prefix convention, plus an entire retired cot domain — none of which have files. They made --check/status.json look inconsistently named.

--reconcile drops every manifest entry whose parquet file is missing (and empties dead domains). Provably safe: only removes bookkeeping for files that don't exist. On the current store it prunes 44 cot (whole domain) + 44/24/20 bare cot_legacy/disagg/tff ghosts, leaving only the real prefixed entries.