[jhy]

This follows the XML namespace scope optimization in #2556.

XmlTreeBuilder and W3Dom.W3CBuilder both need to track namespace bindings as elements open and close. This extracts that work into a shared NamespaceBindings helper instead of maintaining two implementations.

The helper keeps one active bindings map and records only the changes needed to restore each parent scope. This preserves the XML parser’s improved memory use and processing time while making the same scoped lookup available to the W3C converter.

W3CDom uses one tracker for source bindings and another for declarations already emitted into the W3C tree. That replaces repeated ancestor searches with direct lookups and avoids emitting unnecessary declarations.

W3C DOM fixes

This also fixes namespace conversion in several cases:

• Namespace declarations and prefixed attributes now carry the correct namespace URI, so namespace-aware DOM lookups work as expected. This implements the prefixed-attribute handling discussed in #1839.
• Attributes added after parsing, or included through subtree conversion, now use inherited prefix declarations.
• Namespace declarations now apply regardless of attribute order, and an empty declaration shadows an inherited binding only within its scope.
• With namespace awareness disabled, inherited and undeclared prefixes receive the declarations needed for XML serialization. This preserves and extends the undeclared-prefix behavior from #2087.
• Valid HTML names that are not XML QNames are normalized before being passed to the W3C DOM. For example, a:b:c becomes a:b_c.
• Attributes that the W3C DOM still cannot represent are skipped without dropping their containing element.
• Elements that cannot be represented fall back to text without moving their children or following siblings outside the surrounding tree.