Optimize selector matching and clarify matcher contracts

[andrewiggins]

Summary

• Avoid allocating the remaining selector-parts array when matching the terminal part, while preserving ancestor/sibling search behavior.
• Precompute ASCII-lowercased HTML comparison names for CSS element and attribute selectors instead of normalizing for every candidate.
• Read the fixed id and class attributes directly without redundant name normalization, and inline the remaining attribute lookup.
• Define discriminated matcher contracts with type coverage, keeping one parser construction site and a minimal parser diff.

A shared htmlName property holds precomputed HTML names and is required for element and attribute matchers. name preserves the original spelling for foreign-element matching, while value retains the selector payload. Parser construction uses the property order type, name, htmlName, value. Matcher arrays and NodeList query results are unchanged.

This is the lower layer of the stack. It leaves getElementsByTagName() and its tests unchanged; #720 adds qualified-name matching and reuses the optimized engine for that lookup.

Validation

• Full polyfill unit suite: 428 tests passed across 30 files.
• Polyfill TypeScript check passed.
• Prettier and diff checks passed.

_{Stack created with GitHub Stacks CLI • Give Feedback 💬}