Honor the value attribute on <li> in ordered lists

[gaoflow]

Summary

markdownify numbers <ol> items purely by position, so an explicit value attribute on an <li> is silently ignored and the converted Markdown renders different numbering than the source HTML:

>>> from markdownify import markdownify as md
>>> md('<ol><li>a</li><li value="5">b</li><li>c</li></ol>')
'\n\n1. a\n2. b\n3. c\n'    # expected '1. a', '5. b', '6. c'

Per the HTML spec, value sets that item's ordinal and the following items continue counting from it. The library already honors <ol start>, but never read <li value>.

Fix

In convert_li, before falling back to the positional number, walk from the current item backwards (nearest first, including the item itself) for the closest numeric value; the ordinal is int(value) + offset, where offset is the number of items between it and the current one. Non-numeric or negative value falls back to positional numbering, mirroring the existing start guard (.isnumeric()).

number = start + len(el.find_previous_siblings('li'))
for offset, li in enumerate([el] + el.find_previous_siblings('li')):
    value = li.get("value")
    if value and str(value).isnumeric():
        number = int(value) + offset
        break
bullet = '%s.' % number

This composes correctly with <ol start> and with multiple value attributes:

HTML	before	after
<ol><li>a<li value="5">b<li>c	1, 2, 3	1, 5, 6
<ol start="2"><li>a<li value="10">b<li value="4">c<li>d	2, 3, 4, 5	2, 10, 4, 5
<ol><li value="foo">a<li>b	1, 2	1, 2 (fallback)

Verification

• Added 5 assertions to tests/test_lists.py::test_ol (the bug case, value-on-first-item, the compound start + multiple value case, and non-numeric/negative fallback). They fail before the fix and pass after.
• Full suite: 83 passed. flake8 --ignore=E501,W503 markdownify tests clean.

This pull request was prepared with the assistance of AI, under my direction and review.