README: reposition as learning-lab exhibit (patterns, operator vectors, DeepWiki)

[lentago-claude-runner]

Summary

Repositions this README around the Lentago Labs team-learning-lab framing while keeping every existing technical/editorial section intact.

• Adds ## 📚 Ask this codebase (DeepWiki), placed high, pointing at the DeepWiki-indexed wiki for this repo with three good-first-questions.
• Adds ## 🧭 What this repo demonstrates, an 8-row pattern table (prompt-as-code versioning, eval/regression-gate discipline, thin shared-workflow CI wrappers, required-status-check ruleset, issue-linked roadmap, generated brand header, codified-but-disabled automation, AI-authorship disclosure) — every row links to a concrete file or merged PR in this repo.
• Adds ## 🛠️ Make a change yourself, three operator vectors (ship a prompt version + eval baseline, adopt a fleet-standard CI check, toggle automation off without deleting it), each with the exact edit → PR → merge flow and links to the merged PRs that demonstrate it (feat(prompt): v6 auditor — journal legitimacy (H10) and scoring formula fix #44, feat(baseline): v6 production baseline — H10 functional check, regression gate, scoring calibration #46, Adopt the shared docs-check workflow #52, Disable automated Claude PR review (manual-only trigger) #47).
• Adds the standard Lentago Labs footer block linking the org profile and DeepWiki.
• Lightly condenses the "Input Formats" and "Testing" prose (folded into adjacent sections/bullets) to remove duplication introduced by the new pattern table — no operational detail was dropped.

Unchanged: the generated brand header/banner/badges (owned by lentago/.github → brand/generate.py), the authorship-disclosure paragraph, the ten-heuristic table, the scoring formula, the issue-linked Roadmap, Development Methodology, Repository Structure, License, and Credits.

Part of the fleet-wide repositioning tracked in lentago/.github#88

Test plan

• Verified every relative markdown link in the new sections resolves to a real path in this checkout (workflow files, prompt files, test-sets, CLAUDE.md, LICENSE).
• Confirmed the brand header/banner/badge block is byte-for-byte unchanged (diff starts after line 13).
• Cross-checked all cited PR numbers (feat(prompt): v6 auditor — journal legitimacy (H10) and scoring formula fix #44, feat(baseline): v6 production baseline — H10 functional check, regression gate, scoring calibration #46, Disable automated Claude PR review (manual-only trigger) #47, Adopt the shared docs-check workflow #52) and issue links ([v4] Implement temporal impossibility checks #6, [v4] Implement predatory journal flagging #7, Scoring formula: D×3 base cost makes the headline score misleading for large reference lists #43, etc.) against gh pr list/existing README content.
• docs-check required status check (runs automatically on this PR).

Observed, not changed

• No open PRs existed at analysis time, so all "Proof this works" links in the new operator-vector section point to merged history rather than in-flight examples.