feat(gcp-to-aws): Design summary card + cost baseline-quality labels

[herosjourney]

Summary

The problem, in plain English: Two related blind spots. First, the Design phase does the most interesting work in the whole flow — mapping every GCP service to an AWS target with reasoning — and then tells the user nothing but "AWS Architecture designed. Proceeding to Phase 4." The user's first look at their own architecture comes buried in a report two phases later. Second, cost comparisons can silently mix baselines that measure different things: a GCP figure estimated from Terraform resource configs only captures standing charges for discovered resources, while a user's stated spend band covers their whole bill. Put side by side without labels, a "$10/mo GCP vs $112/mo AWS" row can appear in the same report as a "$1K–$5K/mo" spend band, and the savings math looks broken or dishonest — the exact failure observed in a real test migration.

Changes:

• Design summary card (design.md): after the outer-run handoff gate passes, present a compact chat card — top 3–5 GCP → AWS mappings with the existing user-facing vocabulary (Standard pairing / Tailored to your setup / Estimated from billing only), deferred-specialist services called out, and one line on what Estimate answers next. Skipped on inner workshop reprices; phase-status protocol untouched.
• Baseline-quality labels (estimate-infra.md Part 1): display-label table derived from the existing current_costs.source enum (no new schema field): "Measured from your GCP billing (±5%)" / "Estimated from resource configs (±20–30%, standing charges only)" / "Your stated spend band" / etc. The Estimate chat summary's GCP-vs-AWS line now carries the label.
• Not-comparable rule (estimate-infra.md + generate-artifacts-report.md): an inventory-only GCP figure and a user spend band must never sit side-by-side unlabeled; when both exist and disagree beyond the accuracy band, show both with an explicit "these measure different things" line and name the decision baseline. Report Section 3 requires the baseline badge on the GCP column.

Why

Usability review of the outputs: mid-flow value checkpoints and honest cost comparability were the two report-side gaps. The design card reuses vocabulary and data that already exist (design-refs/fast-path.md, design artifact rationale); the baseline labels reuse the existing current_costs.source enum rather than adding a parallel field.

Test plan

• mise run lint:md — 0 errors
• No schema change: current_costs.source enum reused as-is (schema-estimate-infra.md untouched)
• Inner-reprice paths excluded from the design card (explicit skip), phase-status update order preserved
• Manual: inventory-only run (no billing, Q3 answered) shows both figures with the not-comparable line rather than a single misleading comparison row

Out of scope

• Decision outcome fields (PR feat(gcp-to-aws): decision outcomes (go/conditional/defer/stay) + executive trust panel #173 — the card intentionally doesn't render outcome, which doesn't exist until Estimate)
• Workshop compare table (next PR)
• Labor/professional-services dollar estimates (existing policy: never invented)

Note for maintainers: this PR and #169 both touch the design.md completion area (breadcrumb line vs. card insertion above it). Merge #169 first; this branch rebases cleanly — the card block sits above the line #169 rewords.