PR #TBD — Topology Discovery Finally Reaches Confirm

The Broader Goal at the End of the Previous PR

PR #2311 locked the onboarding-engine wiring design for gh:#2310: five modules from the 2026-10-01 fleet dispatch were built and unit-tested, but never imported by a live command path. The goal after that spec landed was to wire each module into a real entry point, with integration tests that hit CLI/wizard/Vizor rather than the module in isolation.

Strategic Shifts in This PR

No redesign. confirm_topology() already did the right thing as a pure persist step — it expects a topology dict and writes it. The missing piece was the discovery engine that was supposed to produce that dict. This PR takes the plan's second integration approach: call topology_discovery before confirm_topology and feed the scored candidate in.

What This PR Shipped

• propose_topology() and discover_and_confirm_topology() in onboarding_state.py
• Vizor POST /w/<slug>/api/onboarding/topology/confirm runs discovery when the body is empty or sets discover: true, then confirms the candidate; explicit topologies still win
• Wizard _advance_wizard_onboarding_state advances through S4 with the same discover→confirm path
• Integration test on the real VizorHandler.do_POST entry point in tests/test_onboarding_topology.py

Brainstorm Visuals

None for this wiring task.

Progress Toward Full Autonomous Multi-Agent Dispatch

Closing one more "tests green, callers zero" gap from the 2026-10-01 onboarding fleet. Dual-surface onboarding (TUI + Vizor) now actually uses evidence-scored archetype discovery before topology confirmation.

New Goalpost

Remaining gh:#2310 tasks: probe --deps, fleet list/apply, upgrade --all, and the first_win_loop.py wire-or-retire decision (Task 4, sequenced after this PR to avoid wizard/server merge conflicts).