Harden parser input validation and stream resource contracts

[Ismail-elkorchi]

Summary

• validate caller-built trees and traversal arguments before work, including graph ownership, namespaces, names, attributes, templates, and cycles
• separate mandatory BOM detection from optional meta prescanning, commit stream decoding when encoding evidence is final, and avoid retaining decoded byte input unless requested
• return deterministic resource evidence for eager tokenization and fragment parsing, including enforceable step budgets
• make registry audits fail closed, retain reviewed per-case browser differences, and qualify Node, Deno, and Bun agreement on Linux, macOS, and Windows

Breaking changes

• parseFragment() returns a result containing tree and metadata
• tokenizeByteStreamEager() returns a result containing tokens and metadata
• malformed caller-built node graphs and invalid query arguments now fail with HtmlConfigurationError

No deprecated overloads or compatibility paths are retained.

Validation

• npm run qualification:release
• 202 behavior/tooling tests and 89 engine tests
• 3,828 WPT tree executions and 6,998 tokenizer fixtures
• 1,724 document cases across Chromium, Firefox, and WebKit
• 600 fuzz runs and 113/113 mutation kills
• package, supply-chain, resource, performance, and Node/Deno/Bun agreement gates