Refine taro2rn skill: split KNOWN_ISSUES, add when_to_use/allowed-tools, fix core/ path bug (#5)

• Split 20KB KNOWN_ISSUES.md into 4 sub-domain files for progressive disclosure (dependencies / styles / api / monorepo); keep KNOWN_ISSUES.md as a slim index to preserve inbound links from SKILL.md, TRANSFORM_RULES.md, and existing in-the-wild taro2rnTODO.md files.
• Rewrite SKILL.md frontmatter: introduce official when_to_use field for trigger phrases (4 -> 8 variants, incl. en/zh), add Do-NOT-use negative list, declare Taro 3.x / RN 0.72+ compatibility, add allowed-tools (Read Write Edit Glob Grep) for permission-prompt-free UX, embed ultrathink at the design-width / rpx / monorepo decision point only (not globally) for targeted reasoning boost.
• Fix latent bug: 3 broken core/ paths in assets/TODO_TEMPLATE.md (actual dir is references/); generated TODO files were giving users 404 paths.
• Update TRANSFORM_RULES.md known-issues row to point at the 4 split files.

Deliberately rejected (vs external review):

• disable-model-invocation: kills auto-trigger UX for “转 RN” phrases
• paths: review-suggested globs (**/package.json) match every JS project
• effort: high (global): wasteful vs targeted ultrathink keyword
• configurable TODO filename: over-engineering, taro2rnTODO.md is unique enough
• evals/ framework: not officially required, premature for v1