fix: decode URI-encoded local link paths

[andyhrc]

Summary

• Decode URI-encoded local paths before resolving them against the current Markdown file
• Apply the same path handling to internal document links and local images
• Preserve malformed URI escapes from raw HTML instead of aborting link rewriting

Problem

MarkdownIt percent-encodes non-ASCII link destinations. For example, ./目录/中文文件.md#中文标题 is rendered with %E7... path segments. The existing code joined those encoded segments directly into the filesystem path, so readTextFile looked for a literal percent-encoded filename and navigation failed.

The fragment is split before decoding, so encoded heading IDs keep their existing behavior.

Validation

• pnpm lint
• pnpm build
• Manually verified encoded Chinese directories and filenames resolve to their Unicode filesystem paths
• Manually verified escaped spaces/filename characters and malformed % sequences