feat: Lazy-load @napi-rs/canvas to fix Jest open handle warnings

[JonRC]

Summary

This PR implements lazy-loading for the @napi-rs/canvas module to prevent Jest from reporting "CustomGC" open handle warnings when using pdf-parse for text-only operations.

Problem

When importing pdf-parse in a Jest test environment, Jest's --detectOpenHandles flag reports a "CustomGC" open handle. This occurs because:

1. pdfjs-dist automatically loads @napi-rs/canvas in Node.js environments
2. The canvas module creates a native GC (garbage collection) handle
3. This handle persists for the process lifetime, triggering Jest's warning

This affects users who only need text extraction (getText(), getInfo()) but still see canvas-related warnings in their test output.

Solution

Introduced LazyCanvasFactory that defers loading of @napi-rs/canvas until canvas operations are actually needed:

• New file: src/pdf-parse/LazyCanvasFactory.ts - Factory that lazy-loads the canvas module on first use
• Modified: src/pdf-parse/PDFParse.ts - Uses LazyCanvasFactory as default CanvasFactory in Node.js
• Modified: src/pdf-parse/index.ts - Exports new utilities for explicit control
• New tests: tests/unit/test-cleanup/lazy-canvas.test.ts - Verifies lazy loading behavior

Key Changes

1. Lazy Loading: Canvas module is only loaded when getImage() or getScreenshot() is called
2. Cleanup: destroy() now calls clearCanvasCache() if canvas was loaded
3. Exports: Users can access isCanvasLoaded() and clearCanvasCache() for explicit control
4. Backward Compatible: Users can still pass custom CanvasFactory as before

Files Changed

File	Change
src/pdf-parse/LazyCanvasFactory.ts	New lazy-loading canvas factory
src/pdf-parse/PDFParse.ts	Use LazyCanvasFactory as default
src/pdf-parse/index.ts	Export new utilities
tests/unit/test-cleanup/lazy-canvas.test.ts	New tests

Test Plan

• All existing unit tests pass (105 tests)
• New lazy-canvas tests pass (4 tests)
• getImage() works correctly with lazy-loaded canvas
• getScreenshot() works correctly with lazy-loaded canvas
• getText() works without loading canvas module
• Code passes Biome linting
• Manual verification in Jest project with --detectOpenHandles

[JonRC]

@mehmet-kozan The CI checks are failing because fork PRs don't have access to repository secrets needed for the private submodule checkout. Could you please manually re-run the failed workflows?

[JonRC]

@mehmet-kozan, also let me know if this improvement makes sense.

[mehmet-kozan]

@JonRC thanks for your contribution. I was able to reproduce the issue using the jest --detectOpenHandles flag. There is already an option for a custom canvas factory, see:

• https://github.com/mehmet-kozan/pdf-parse/blob/main/docs/options.md#load-parameters
• https://github.com/mehmet-kozan/pdf-parse/blob/main/docs/troubleshooting.md

We can add a integration test for jest --detectOpenHandles flag tod tests/integration dir.
Node.js-related code must be placed in the pdf-parse/worker or pdf-parse/node submodule for browser builds. We can add the LazyCanvasFactory implementation to one of these submodules.

I will also add you permanently to the pro package repository, see: #77

We can rework this feature together in another branch if you’d like. This would also allow me to test the contribution workflow.