Don't raise from parse() on entity backslashes, out-of-range character references, or non-numeric EPSG codes

[DrVelvetFog]

Fixes three of the cases in #599:

• sanitizer.replace_doctype(): pass the rebuilt entity declarations to RE_DOCTYPE_PATTERN.sub() through a function, so backslashes in entity values aren't read as escapes or group references. Before, C:\data raised re.PatternError, C:\temp turned into a tab, and \1 pulled in DOCTYPE text.
• mixin.handle_charref(): map character references above U+10FFFF, and surrogates, to U+FFFD (what HTML parsers do) instead of raising from chr() / .encode().
• namespaces/georss.py: if the EPSG code in srsName isn't an integer, treat it like a missing srsName instead of raising from int(). The check that was duplicated in _end_gml_pos and _end_gml_poslist is now one helper.

Tests, in the existing data-driven format:

• tests/wellformed/rss/entity_in_doctype_backslash.xml: C:\data\1\new comes through unchanged
• tests/illformed/charref_out_of_range.xml: �, � and � become U+FFFD
• tests/wellformed/geo/gml_point_epsg_not_numeric.xml: srsName="EPSG:unknown" parses like a point with no srsName

All three fail without the change (5 failures across the strict and loose parsers). The full suite passes: 4302 passed, 8 skipped. black and isort are clean. Changelog fragment added under "Fixed".

[kurtmckee]

This seems to be generated by AI. Closing.