Implement one-shot setup UX

[minsing-jin]

Summary

• Add a one-page /setup flow that makes OPENAI_API_KEY the only required first-run input, defaults to local SQLite, generates criteria from optional interest/defaults, initializes source defaults, and can trigger the first feed run from setup.
• Preserve existing features by layering them behind advanced/progressive sections or links: Supabase, delivery, source API keys, export/import, profile/status/settings, model/backend settings, onboarding, feed actions, and advanced algorithm routes.
• Add route inventory documentation and regression coverage for setup, onboarding route independence, local SQLite schema, one-shot feed execution, source defaults, and preserved user-facing paths.

Ouroboros Process

• Interview: interview_20260518_022059, ambiguity 0.1185.
• Seed: original seed_ee34c8c478c3; recovery regeneration from the same interview produced seed_520639e6c69f; both were treated with the explicit no-feature-removal constraint.
• Run: job_881e6eecc7de / orch_d7d27d10ff54 / exec_3c4140a1fa31, completed AC 15/15, Sub-AC 159/159.
• Eval: official ouroboros_evaluate succeeds when addressed to execution ID exec_3c4140a1fa31: Final Approval APPROVED, Stage 2 completed, score 0.82, AC compliance YES, drift 0.18. Earlier calls against orchestrator session orch_d7d27d10ff54 timed out at 120s. Bounded fallback ouroboros_qa also passed the concrete evidence artifact at 0.94 with threshold 0.90 (qa-hedwig-issue32-recovery-20260519).
• Evolve: existing lineage lin_hedwig_issue32_one_shot_setup_ux_nf_20260518 remains Gen 1: completed | pending; official recovery attempts with valid seed content still timed out or stalled (job_7fdcf6dfbb09, job_fdb69a3db266, job_f82197daada7, plus sync lineage lin_hedwig_issue32_official_recovery_20260519_03 at Generations: 0) and were cancelled when cursor/updated timestamps stayed fixed.
• Ralph: prior low-resource serial attempts reached Generation 1 completed but did not terminalize. Latest official attempts also failed to terminalize: job_4f00b59c2142 (execute=false) and job_361cd33b34cf (execute=true, parallel=false, max_generations=1, max_total_seconds=900) stayed at Running ralph with unchanged cursor/updated timestamps past their configured timeouts, then were cancelled to avoid orphan/OOM-prone background jobs. This appears to be an Ouroboros MCP job terminalization/runtime issue, not an implementation diff failure.

Verification

• git diff --check
• Focused suites: 170 passed
• Full suite: 861 passed, 1 warning after adding a temporary local .venv symlink for tests that invoke .venv/bin/python; symlink was removed before commit

Closes #32