fix: URL-encode rule doc link targets

[bmish]

Summary

• URL-encode each path segment in generated relative rule-doc links (pathToUrl), including characters encodeURIComponent leaves alone that break CommonMark destinations ((, ), !)
• Encode only the substituted {name} portion of string --url-rule-doc templates; leave the template itself untouched
• Encode relative urlRuleDoc function return values; leave absolute URLs (://) as-is so callers own encoding

Test plan

• npm run lint
• npm test
• New fixtures with rule name/path no foo (bar) assert exact %20 / %28 / %29 in snapshots for relative links, string templates, and relative function returns
• Absolute function return with pre-encoded URL is not double-encoded

Made with Cursor

Context

From the post-#987 standards audit: pathToUrl handled Windows separator conversion but never URL-encoded, so a rule doc path containing a space or parenthesis produced a broken link — CommonMark terminates a plain link destination at whitespace or an unbalanced ). (, ), and ! also pass through encodeURIComponent untouched, hence the explicit replacements.

Scope guard: only tool-generated relative paths and the substituted {name} portion are encoded; user-supplied absolute URLs and template text are left to the caller, so nothing pre-encoded gets double-encoded.