feat(scan): add stats.regime.cusum_break@1 CUSUM structural-break detector

[radiusred-cody]

stats.regime.cusum_break@1 — CUSUM structural-break / regime detection

New single-leg ANOM scan that detects structural breaks in the mean (drift) and/or volatility of the log-return series, and surfaces them as a meta-signal usable as a conditioning filter on other signals.

Method

• Mean — standardized CUSUM (sup of a Brownian bridge; Kolmogorov critical value).
• Vol — Inclán & Tiao (1994) CUSUM-of-squares.
• Multiple breaks — "Bai-Perron-lite" binary segmentation; a segment is only split when its statistic exceeds threshold and both children are ≥ min_segment.

Surface

• id = stats.regime.cusum_break, version = 1, arity Single. Registered alphabetically in register_anom_scans (Pattern E — registry.rs untouched); dispatchable on CLI/MCP/HTTP via the registry.
• params: target (mean/vol/both, default both), threshold (default 1.358, lower = more sensitive), min_segment (default 30).
• Emits one located Finding::Result per detected break (break index/timestamp, pre/post segment mean/std/median, regime ordinal, target). A stationary series flags none.

Reuse (no duplication)

• Factored the median order-statistic helper out of the cointegration_rolling kernel into a shared scan/primitives/robust module, consumed by both that scan's beta-drift baseline and this scan's robust pre/post segment center. Cointegration behaviour is byte-identical — its existing integration test is unchanged.

Tests

• Unit tests (kernel + surface) + integration test tests/scan_cusum_break.rs with an insta schema snapshot. Acceptance: injected mean / vol shift flags a break near the true changepoint; a stationary series flags none; min_segment respected; deterministic.

🤖 Generated with Claude Code

[radiusred-testy]

Approved

Verdict: QA PASS — all 4 acceptance criteria met.

CI

Two runs triggered. Run 27068176043 (push-triggered, completed at 17:02:47 UTC) — ✅ SUCCESS — all steps green: cargo build, cargo clippy -- -D warnings, cargo fmt --check, cargo test, tokio-free miner-core, schema sync, cargo audit, cargo deny check.

Run 27068187094 (PR-event-triggered) — infrastructure failure only: System.IO.IOException: No space left on device in the runner log daemon after all code-quality steps passed (cargo build, cargo clippy, cargo fmt --check, cargo test, tokio-free, schema sync` all ✓). Disk exhaustion is a runner environment issue, not a code defect.

Conventional commit

feat(scan): add stats.regime.cusum_break@1 CUSUM structural-break det… — correct type for a new user-visible scan.

AC checklist

• AC1 — registered + dispatchable: CusumBreakScan registered alphabetically in register_anom_scans (Pattern E, anom/mod.rs:1267); is_some() assertion added in the family's mod.rs tests; auto-dispatches on CLI/MCP/HTTP via registry. ✅
• AC2 — synthetic series: injected mean shift (0.05 drift on 0.01-scale noise) and vol shift (8× volatility) both flagged ≤20 bars from the true changepoint at kernel and integration level; stationary series with conservative threshold 2.5 flags none. ✅
• AC3 — shared primitive: median_in_place relocated from cointegration_rolling/kernel.rs to scan/primitives/robust.rs; algorithm byte-identical; both scans import from the shared module; tests/scan_cointegration_rolling.rs untouched. ✅
• AC4 — integration test + insta snapshot: crates/miner-core/tests/scan_cusum_break.rs + snapshots/scan_cusum_break__cusum_break_schema.snap added; nextest + clippy green on successful CI run. ✅

No issues found

No security issues, no private research data, no logic errors, no out-of-scope changes.

Ready to merge. Merge with --rebase --delete-branch as requested.