Twitter extractor: hardcoded English aria-label breaks reply extraction for non-English interfaces

[nanma]

Bug

TwitterExtractor uses a hardcoded English aria-label to locate the conversation timeline:

// src/extractors/twitter.ts line 16
const timeline = document.querySelector('[aria-label="Timeline: Conversation"]');

When the X/Twitter interface is set to a non-English language, the aria-label is localized (e.g. `"时间线：对话"` in Simplified Chinese), so `timeline` is always `null`. The extractor then falls back to finding a single `article[data-testid="tweet"]`, which locates the main tweet but leaves `replyTweets` empty — so replies are silently dropped even though the DOM is fully loaded.

To reproduce

1. Set X/Twitter display language to any non-English language (e.g. Simplified Chinese)
2. Open a tweet with replies loaded in the DOM
3. Clip with Obsidian Web Clipper — only the main tweet is captured, replies are missing

Suggested fix

Match the container by a stable, language-independent attribute instead. Two options:

Option A — match on a substring common to all locales (the timeline container always contains cellInnerDiv children):

const timeline = Array.from(
    document.querySelectorAll('[aria-label]')
).find(el => el.querySelector('[data-testid="cellInnerDiv"]') !== null) ?? null;

Option B — partial aria-label match using a known locale map (more explicit, easier to extend):

A mapping of known localized values could be maintained, or a looser selector like [aria-label*="Conversation"] combined with a structural check.

Option A is simpler and doesn't rely on any string content at all.

Environment

• defuddle version: 0.18.1
• Observed with: Simplified Chinese (zh) interface on x.com

[mvanhorn]

Implemented in #290 - swapped the localized aria-label match for a data-testid lookup which X uses consistently across locales.