fix(html): keep caption, colgroup and tfoot from breaking table headers

[om singhal (Om-singhaI)]

Fixes #2536

markdownify only treats a <tr> as the first row when it has no previous sibling. html.parser doesn't add a <tbody>, so when rows sit directly under <table>, a <caption>, <colgroup> or <col> before them drops the header's | --- | line. A <tfoot> row has no previous sibling either, so it's rendered as a header: a <td> footer gets an empty header and delimiter, and a <th> footer gets a delimiter under it.

The check lives in markdownify's convert_tr, so I fixed it in _CustomMarkdownify. A new _normalize_table_sections helper runs first in convert_soup:

• A direct <caption> moves in front of its table, and a convert_caption override adds a blank line before it so it doesn't glue onto text before it.
• <colgroup> is unwrapped and <col> dropped, since html.parser can leave an unclosed <colgroup> holding every row.
• <tfoot> rows move to the end of the last <tbody> (or the table), which also puts HTML4 style footers last.

Testing:

• 10 new cases in tests/test_html_converter.py fail on main and pass here.
• The related test files pass (test_html_converter, test_module_vectors, test_module_misc, RSS, EPUB, Office images). Skips need network, an API key or exiftool, and 2 test_docintel_html failures happen on main too.
• All of tests/test_files converts to identical output before and after.
• black 23.7.0 is clean.