fix(html): extract definition lists instead of discarding them

[r0h1tb]

Problem

<dl>, <dt> and <dd> are mapped to RemovedBlock, so partition_html() drops definition lists and everything inside them. Documentation gets hit hardest: Sphinx renders every documented function, class and attribute as a <dl>, so partitioning an API reference page returns the headings and nothing else.

partition_html(text=(
    '<h1>API reference</h1><dl class="py function"><dt>connect(host, port=5432)</dt>'
    "<dd><p>Open a connection to the database server.</p></dd></dl>"
))
# [Title('API reference')]

Glossaries and key/value lists disappear the same way, with no error.

Fix

Map <dl> to ListBlock and <dd> to ListItemBlock, which the parser already anticipates: the ListBlock docstring says "maybe a <dl> element at some point", and the list-depth code already counts dl ancestors for dd items. <dt> becomes a plain BlockItem, so a term gets its type from its text like any paragraph. The v2 (ontology) parser already keeps definition lists.

Tests

Two new tests in test_partition.py: a glossary (exact element types, text and category_depth) and a nested Sphinx-style API reference (all text kept, in order). Without the fix they return [('Title', 'Glossary', 1)] and ['API reference']. The html, chunking, documents, md, text and email tests go from 1201 passed on main to 1203, with no failures.

This touches the same mapping table as #4451 (<details>/<summary>), so whichever lands second will need a small rebase.

[cragwolfe]

SAFE TO MERGE at reviewed head 89f382c4799a1c0d991123d1141bdab879f5f044 against main at 1bedf7be0bea9db5c5dc3a9a2dab83078b4db54d. No actionable Critical or High findings remain.

The changed tags reuse existing traversal classes; the parser already counts <dl> ancestors for <dd> depth. The added tests cover exact glossary element types and depth, plus nested Sphinx text retention in source order. All six required exact-head checks passed at review time.

(authored by codex)