Note: Although lxml.html.clean is no longer part of lxml, defs.safe_attrs, which is involved with the problem, still seems to be in lxml.

## The issue
Currently, lxml.html.clean.Cleaner strips all aria-* attributes (e.g., aria-label, aria-hidden, aria-describedby) and the role attribute by default.

This behavior causes sanitized HTML to lose semantic meaning for assistive technologies (screen readers), making the output non-compliant with WCAG standards.

## Reproduction
“`py
from lxml.html.clean import Cleaner

cleaner = Cleaner() # Default settings
html_input = '<div role="alert" aria-live="assertive">Error message</div>'

# Current Behavior:
# Output: <div>Error message</div>
print(cleaner.clean_html(html_input))

# Desired Behavior:
# Output: <div role="alert" aria-live="assertive">Error message</div>
“`

## Fix

I have prepared a patch that adds ARIA attributes to the default allowed attributes list: https://github.com/oomsveta/lxml/commit/7994d4e61efad9a81fc8c32af8c33001bdf93424

This change ensures that developers using lxml for sanitization do not accidentally strip accessibility features while maintaining protection against XSS.

## — PLEASE PROVIDE THE FOLLOWING INFORMATION: —
Python : sys.version_info(major=3, minor=13, micro=5, releaselevel='final', serial=0)
lxml.etree : (5, 2, 1, 0)
libxml used : (2, 9, 14)
libxml compiled : (2, 9, 14)
libxslt used : (1, 1, 35)
libxslt compiled : (1, 1, 35)