Modernize Legends GitHub for portable, evidence-backed repository work

[avalonreset]

The existing suite coupled its guidance to specific agent hosts, treated missing GitHub access as failed checks, and required an image provider for artwork. This refresh makes the command runtime the common interface and gives agents evidence-backed priorities instead of treating a checklist score as the objective.

Changes:

• Add a portable launcher, capability JSON, offline execution, and isolated artifacts; retain legacy entrypoints and score fields.
• Add versioned audit findings, availability, repository profiles, and practical verification steps.
• Add a general-purpose discovery planner for examples, factual comparisons, distribution experiments, and measurement.
• Remove KIE.ai runtime and installer setup; reuse supplied local artwork without credentials.
• Rewrite skills and specialist guidance; remove README mascot imagery and unsupported search-ranking claims.
• Fix audit .gitignore side effects, JSON write integrity, offline mutation guards, and existing-topic preservation.

Validation:

• Python regression suite and Markdown lint locally.
• All eleven planning workflows exercised offline against Legends GeoGrid; Git status and hashes of 74 tracked files unchanged.
• Separate live GeoGrid audit confirms remote evidence collection.
• Installer syntax and installed skill path/help checks.
• CI adds Windows/Linux on Python 3.10/3.12.

Native integrations for every named agent host are not certified. Discovery outputs are plans, not claims of increased traffic. This draft does not replace the published release.

Product acceptance follow-up

GeoGrid output review caught issues that the original smoke tests did not: README layout loss and inconsistent repository classification are now fixed. Existing authored READMEs are preserved for targeted edits. Added an optional local typography renderer matching the owner's approved banner exactly, with no bundled font or provider; social previews retain the complete image. Reconciled the concurrently published banner from main.

Keep this PR in draft: docs/GEOGRID-ACCEPTANCE.md records four remaining specialist-planner issues involving CI recognition, provenance heuristics, changelog parsing and unavailable metadata. Current local suite: 89 tests, 88 passed and one Windows symlink skip. No release was published.

DataForSEO integration

Adds the shared public legends-dataforseo-kit dependency for live research, a bounded estimate-first research command, and validated keyword/SERP import into SEO artifacts. Preserves raw observations, locale, timestamps, and costs; does not treat search demand as measured traffic gains. Replaces legacy MCP installation and credential-file setup with the shared Python package while retaining installer entrypoints. Existing host configuration is left untouched.

GeoGrid pilot: 16 submitted keyword seeds, 13 returned records, three live SERPs, $0.01956 reported cost. See docs/GEOGRID-SEARCH-STRATEGY.md for proposed product copy and evidence interpretation. The toolkit's four specialist acceptance issues remain documented; this PR stays draft.