feat(extractor): impl HwpxExtractor for application/haansofthwpx

[kh3rld]

Summary

• Adds HwpxExtractor for HWPX (Open HWPML / Hangul Word Processor XML) documents
• Uses unhwp crate (MIT, default-features = false, features = ["hwpx"]) — avoids writing custom OWPML XML parsing
• Extracts headings, paragraphs, tables, and embedded images
• Adds ZIP magic-byte detection for application/haansofthwpx in the MIME pipeline
• Gates behind a new hwpx Cargo feature, included in formats and full aggregates
• HwpxExtractor added to alef.toml exclude_types — it is an internal implementation type, not a public API surface

Why unhwp instead of custom XML parsing?

HWPX is a ZIP-of-OWPML format with a non-trivial schema (sections, paragraphs, inline runs, image resources). unhwp models this correctly in ~zero lines of custom parsing and is MIT-licensed with no runtime deps when used with default-features = false.

Test plan

• cargo test -p kreuzberg --features hwpx — unit tests in hwpx.rs
• fixtures/format_specific/format_hwpx_standalone.json fixture added; run alef e2e generate && task test:e2e to verify across all language bindings
• MIME detection: ZIP with Contents/content.hpf marker routes to application/haansofthwpx

Closes #849