v2.4.2

PDF Craft v2.4.2

PDF Craft 2.4.2 fixes OCR-based document metadata extraction and unifies business
execution around async implementations. PDFCraft remains the synchronous facade
over AsyncPDFCraft.

OCR metadata extraction

Book metadata extraction now awaits the guaranteed JSON repair loop and LLM runtime
throughout the request chain. This fixes the blocking guaranteed JSON request returned an awaitable error and the unawaited-coroutine warning that caused extraction to fall
back to native PDF metadata. When a PDF has no native title, a supported title from
the OCR response can now reach document.title in the exported PCEX.

A regression test exercises the actual LLM runtime with a simulated network response
and verifies that Attention Is All You Need reaches the PCEX manifest when native
metadata is empty.

See #503.

Async execution and extension contracts

• LLM requests, JSON/XML repair loops, translation, and executable extensions use
  async contracts. LLM configuration remains declarative; tokenizer and prompt
  resources are loaded through the async runtime.
• PDF patching prepares extraction content and runs Ghostscript asynchronously,
  with native Qt composition isolated in a worker process. Custom PDF handlers and
  visual compilers execute on the caller's event loop.
• Markdown and EPUB rendering use async orchestration. Filesystem operations and
  synchronous third-party generation remain in dedicated execution domains.
• OCR metadata and table-of-contents file scans run outside the event-loop thread.
• Cancellation cleans temporary resources and cancels pending async fill-recovery
  callbacks. Fill-recovery callbacks run on the caller's event loop.

Public API

• Await PDFTranslationPipeline.patch(), PDFPatcher.patch(), and
  GhostscriptVisualBaseCompiler.compile().
• Custom PDF providers implement AsyncPDFHandler and AsyncPDFDocument; custom
  visual compilers implement async def compile(source_path, target_path).
• Model download uses await predownload_models(...) or
  await AsyncPDFCraft().predownload_models(...). Synchronous applications use
  PDFCraft().predownload_models(...).
• The package's public PDF provider contracts are AsyncPDFHandler and
  AsyncPDFDocument. Native synchronous PDF adapters are internal implementation
  details.

The API references and installation guides describe these execution contracts.

Full Changelog: v2.4.1...v2.4.2