feat(wasm): stream XML records and benchmark runtimes

[nth-bailey]

Summary

Completes the next implementation slice of #47 after #69:

• Add parseStream for schema-free record documents. It accepts a web ReadableStream or async iterable, emits each direct child once complete, and bounds retained record size with maxRecordBytes.
• Test byte-by-byte chunks, split UTF-8, quoted tag delimiters, entities, CDATA, namespace prefixes, early emission, record limits, and incomplete documents in Node and Bun.
• Add reproducible micro, medium, and 50 MB benchmark harnesses for Node, Bun, and headless Chromium; compare object output to fast-xml-parser, and include xml2js/DOMParser as separate references.
• Publish the measured results, bundle size, and limitations in the benchmark docs. CI now smoke tests all three runtimes.

Scope and limitations

• Streaming is for repeated direct children of one root and uses schema-free conversion. It buffers one complete child, plus temporary wrapper and Wasm copies. It does not incrementally parse a single huge child, support DTDs or mixed root text, or offer typed XSD streaming.
• Measurements are exploratory on one repeated-record fixture; GC pause comparisons were not reliable, so no GC reduction claim is made.
• This PR does not publish the npm package; that release task is tracked in Publish @polyxml/wasm to npm #71.

Verification

• cargo fmt --check, cargo clippy --workspace --all-targets -- -D warnings, cargo test --workspace -q
• Wasm package build; 8 Node and 8 Bun tests
• Node, Bun, and Chromium benchmark harnesses across small/medium/large tiers
• npm audit clean, npm pack --dry-run includes stream helper
• uvx zensical build --clean --strict

Closes #47