fix(html): correctly handle <caption> element in HTML table deserialization

[Satvik77777]

Fixes #635

Problem

HTML tables with a <caption> element produced broken Markdown output — each cell rendered on its own line instead of a valid table row:

Before (broken): Employee Details
| | Name
| Role
|
| Alice
| Engineer
|

Root Cause

The <caption> element was being processed as a regular table child node, corrupting the TableHead/TableBody structure during deserialization. This caused the table rows to break apart into individual lines.

Fix

• Extract <caption> directly from the DOM before processing table children — no leaking intermediate nodes
• Emit caption as a standalone Paragraph block before the table (plain text, no Strong wrapping of arbitrary content, no \n\n text injection)
• Filter table children to TableHead/TableBody only, ignoring other child nodes
• Promote first body row to header when it contains th cells and no thead exists
• Normalize whitespace in table cells via cleanTableNodes() to remove Softbreak nodes and merge adjacent Text nodes
• Add regression tests covering: caption present, no caption (no regression), whitespace normalization in caption

Note on #640

This PR addresses the same issue as #640 but resolves all review feedback that blocked that PR from merging:

• ✅ Tests included (missing in fix(table): handle HTML table captions correctly #640)
• ✅ Caption extracted from DOM directly — no leaking type: 'caption' nodes
• ✅ No Strong wrapping of arbitrary block content
• ✅ No \n\n text node hack inside Paragraph
• ✅ 4-space indentation throughout (ESLint compliant)

Checklist

• DCO sign-off provided (--signoff)
• Regression tests added
• Commits follow AP conventional commits format
• No documentation changes required
• Merging to main from fork