fix(web): resolve parsed Windows file links - #1

Problem

The upstream PR preserves direct Windows drive links through sanitization, but ChatMarkdown still classifies file links from a raw-text regex before rendering. Valid Markdown destinations such as <C:/Users/Carlos/My Project/docs/Plan (final).md> are parsed by React Markdown but missed by that regex, so the precomputed metadata lookup is empty and the rendered link falls through to the ordinary external-anchor path.

Fix

• Resolve the normalized, parsed href directly when the precomputed raw-Markdown lookup misses.
• Keep the precomputed map as the fast path.
• Add regression coverage for angle-bracketed Windows paths containing spaces and balanced parentheses.

This is intentionally a small follow-up to pingdotgg#6237 and covers the reproduction documented in pingdotgg#5158.

Verification

• vp test run src/markdown-links.test.ts src/markdown-links-rendering.test.tsx --project unit — 43 tests passed
• pnpm --filter @t3tools/web typecheck
• Targeted vp lint and vp fmt --check for the four changed files

No visual styling changes; this only routes already-rendered file links through the existing file-link behavior.

Implemented with GPT-5.6 Sol in T3 Code via the Codex harness.