fix: adjust FxTwitter facet indices for surrogate-pair emoji

[Thinkscape]

Problem

FxTwitter API returns facet indices counting Unicode code points (emoji = 1), but renderTweet() and applyFacets() treat them as UTF-16 code units (emoji in the supplementary multilingual plane are surrogate pairs that count as 2 in JavaScript strings).

When a tweet contains emoji (🦞, 🩺, 🔌, 🌐, etc.) before a URL or mention facet, the facet markers are placed at incorrect positions. This produces broken HTML like:

<p>Small maintenance relea<a href="https://t.co/VkjTYkoy7V">se:<br>...</a>oy7V</p>

After Turndown conversion, this becomes garbled markdown:

Small maintenance relea[se:  
https://t.co/VkjTYk](https://t.co/VkjTYkoy7V)oy7V

Real-world example

Tweet: https://x.com/openclaw/status/2052096219233587451

The tweet contains 4 emoji before the t.co URL, causing a 4-code-unit offset between FxTwitter facet indices and JavaScript string positions.

Fix

Added two helper methods to XOembedExtractor:

• codePointToUtf16Index() — converts a single code-point index to its UTF-16 code-unit equivalent by iterating over characters and counting surrogate pairs
• adjustFacetIndicesToUtf16() — applies the conversion to all facet indices, with a fast-path returning early when the text contains no surrogate pairs

These are called in renderTweet() before facets are passed to applyFacets().

Tests

Added tests/x-oembed-surrogates.test.ts with 3 test cases:

1. Tweet with emoji — uses the real FxTwitter API response for the bug-report tweet, verifies no broken link placement
2. Tweet without emoji — verifies indices pass through unchanged when no surrogate pairs exist
3. Markdown output — verifies Turndown produces clean markdown after the fix

All 284 existing tests continue to pass.