improvement(cli): coherent post-setup flow for success vs. failure

[bjoaquinc]

Overview

The post-setup output didn't make sense on the non-happy paths: a failed run still rendered a "You're all set" panel, the resume paths dead-ended at uv sync, an orphaned 1./2. narrative leaked into paths where the run never happened, and the agent prompt told users to "create your own pipeline" even after a failed run. This reworks the flow so the framing always matches the outcome.

Changes

• Unified the post-setup panel. Merged print_next_steps + print_resume_steps into one builder. Title is outcome-driven: "You're all set" only after a successful run, "Almost there" otherwise (failed run / uv declined / --scaffold-only).
• Resume paths point forward. When setup is unfinished, the panel lists the remaining install steps (Finish setup) followed by the sample-run steps (What to try next) as one ordered list, instead of stopping at uv sync.
• Dropped the stale 1./2. numbering from the playground/agent headlines — it only lined up on the happy path and left an orphaned 2. wherever the run was skipped or failed.
• Outcome-aware agent picker header. choose_agent(ran=...) shows the "create your own pipeline into your playground" lead only after a successful run, and a neutral "set up your coding agent" lead otherwise — so the failed path no longer presumes a working demo or contradicts the "run the sample first" steps.

Tests

• 116 unit tests pass; fast integration tests pass; ruff clean.
• New regression tests assert the picker header swaps its lead on ran true/false; test_display.py updated for the unified builder (incl. that the finish-setup panel still points to the sample run).