fix: update feed list SDUI extraction selectors (#569)

[sigvardt]

Summary

Fixes #569 by updating the feed list extraction logic to correctly handle the current LinkedIn SDUI structure.

Changes

• Removed reliance on componentkey regex matching for legacy parsing and replaced with pure structural DOM inspection for the new SDUI timeline layout.
• Added heuristics to locate profile-displayphoto, company-logo, and ghost-person images.
• Extracted aria-label or alt text from avatar images as primary author name (this resolves the issue where name and headline are merged in the standard text span).
• Added logic to look at adjacent text blocks near the author name to accurately find the headline and posted_at timestamps instead of relying on broken CSS classes like .update-components-actor__sub-description.
• Preserved existing structure for expandable-text-box extraction while adding a new fallback to grab the longest text block when missing.
• Added metric extraction (reactions_count, comments_count, reposts_count) directly from <button> and <span> text without depending on previous CSS selectors.

Closes #569