Expand cascade family to ~12–15 diverse scenarios

[Connorrmcd6]

Expand the cascade scenario family for robustness — breadth, not depth.

The v1 headline rests on just 4 structurally-similar cascade scenarios; at the saturated 0/100 effect, more N won't move it but scenario diversity defends it. Grow to ~12–15 total cascade scenarios spanning distinct drift types, languages, and domains, so reviewers can't attribute the effect to one drift kind or to Python.

Author per bench/scenarios/CHECKLIST.md (new): clone the cascade-quota-batcher-code structure, neutrally-worded task (no worked example leaking the stale value — see #9 lesson), graders that derive ground truth by probing the real hidden dependency (never hardcode T1), then tools/author.py to seal hubs + emit genuine surf_report.json.

Diversity axes to cover:

• drift type: boundary flip (<→<=), constant change, condition inversion (allow→block), dropped guard/await, units (ms↔s), default-arg, off-by-one
• language: Python, TS/JS, + at least one Go or Rust (defends "not a Python artifact")
• domain: rate-limit, retries, auth/access, pagination, cache/TTL, currency/rounding, dates/tz, serialization, concurrency
• ~3 qa (cheap) + rest code; spread across tiers T1–T3
• oracle tripwire per scenario: C2-fresh ≈100% (catches authoring bugs)

[Connorrmcd6]

Merged: 4→14 cascade scenarios on main (#124, Connorrmcd6/surface#126, Connorrmcd6/surface#127). All seal, pass grader-polarization (validate_scenario --all), and the mock multi-mode run is error-free.