Refine adaptive keyword selection and scope image tool

[arthmaker]

Motivation

• Make generated anchor keywords more natural by preferring contiguous phrase windows from titles so anchor text reads in-order and stays contextually coherent.
• Improve scoring consistency and avoid repeated scanning when ranking tokens to produce more stable keyword choices.
• Prevent global/DOM symbol collisions and keep the image link tool usable alongside the anchor/link tool by scoping its behavior and adding defensive guards.

Description

• Rewrote extractAdaptiveKeywords to score title tokens, build a scoreByWord map, and select the highest-scoring contiguous token window (2–4 words) before filling remaining slots from ranked tokens, and ensured presence of mahjongways when present in the title.
• Tuned TOKEN_BOOST and expanded STOPWORDS_ID to favor contextual/tech terms and avoid noisy tokens when generating anchors.
• Merged image UI into anchor.html, removed standalone image.html, and included image.js while updating copy and labels for combined Anchor/Link/Image workflow.
• Scoped image.js inside an IIFE, replaced the global $ helper with getEl, added defensive ?. checks and existence guards, preserved auto-save/load behavior, and adjusted style.css and index.html layout (single-column grid, .three sizing, .stack, .toggle-row, dropzone file-button hiding, and action/button styles).

Testing

• No automated tests were executed for this rollout; the changes are logic/UI focused and were not subject to automated test runs in this run.

---

Codex Task