fix(phase6): unconditional post-Learn capture + ban cold-boot on Deliver failure (#618 iter)

[jjackson]

Iteration on #717's #618 probe. Run 20260605-2138 exposed the deeper bug: on a Deliver-leg failure the Phase-6 agent ran mobile_ensure_avd_running (cold-boot) to "recover" — which wiped the post-Learn state and reset the device PIN, wedging the rest of the run on the Android auth prompt (com.android.systemui). The agent's post-failure dump was the PIN screen, not the CommCare home — so the probe captured nothing, and this cold-boot is the engine behind #618's per-run thrashing (failure → cold-boot → state loss → re-register → give up).

Two fixes in app-screenshot-capture:

1. Capture the post-Learn landing unconditionally at Deliver-leg start (journey-deliver/00-postlearn-landing.xml), before any Deliver recipe or remediation — so the Phase 6 Deliver leg: journey-learn lands in CommCare home, connect-resume-opp asserts Connect jobs-list visible — inter-leg navigation gap #618 ground truth is captured no matter what fails next. Classifies the pathway (nsv_home_screen=Path A vs connect_fragment_jobs_list=Path B).
2. Never cold-boot / re-bootstrap to recover a Deliver-leg failure. A Deliver landing on the CommCare home is the Phase 6 Deliver leg: journey-learn lands in CommCare home, connect-resume-opp asserts Connect jobs-list visible — inter-leg navigation gap #618 nav gap (expected), not a device-infra failure; cold-boot resets the PIN and blocks recovery. Capture + record incomplete + STOP. Re-bootstrap stays reserved for the Learn-leg PersonalID-wipe row only.

The logout probe now triggers off the unconditional Path-A detection (not a connect-resume-opp-specific failure, which didn't match the agent running journey-deliver.yaml). Full suite green.