Add cgis init-ontology: auto-propose patterns.yaml from measured graph

[zaebee]

Summary

Add cgis init-ontology — auto-propose a starter patterns.yaml from the measured graph, so cgis_drift is usable without hand-authoring the ontology.

Motivation

cgis_drift is the most valuable feature but has the steepest onboarding cost: it needs a hand-written patterns.yaml binding each domain (fqn_prefix) to an expected_pattern + drift_tolerance. On a fresh repo (drift on a clone is a no-op without it), this is the single biggest barrier to adoption.

When I onboarded a real FastAPI backend I did this manually and mechanically:

1. enumerate top-level packages → candidate domains (app.services, app.crud, app.utils, …)
2. measure each domain's actual triad census
3. pick the closest of the 5 templates as expected_pattern
4. set drift_tolerance = measured + small margin (ratchet baseline)

Steps 2–4 are exactly what the engine already computes. It can propose the file.

Proposal

cgis init-ontology(db_path, out="patterns.yaml", margin=0.03, min_nodes=10)

• discover candidate domains from package/module structure (configurable depth)
• for each, compute the census and pick the nearest template by TV distance (report the fit + the runner-up)
• emit drift_tolerance = measured + margin and a # measured ≈ X comment per domain
• domains below min_nodes → hygiene-only (no expected_pattern)
• reuse the existing language profiles / patterns blocks verbatim

Output is a ready-to-edit baseline the user ratchets down over time, instead of a blank page.

Impact

Turns drift from "expert-only, write YAML first" into "run one command, get a baseline." This is the difference between a demo and something teams actually adopt in CI.