Use our fork of `skia-canvas` for rendering which handles
encoding characters correctly with the caveat font and other
similar fonts that can group glyphs like ligatures.
This resolves issues with pdf text extraction where characters
were unable to be extracted due to lacking any data within the cmaps.
Seems there's a memory leak in gpu rendering with skia canvas
where contexts can live for much longer than expected escaping gc
cleanup
CPU rendering seems better albeit a bit slower.
Synthetic tests were ran with `--expose-gc` to simulate load over time.
Add proper cleanup for PDF.js pages and loading task to prevent memory
leaks when
processing multiple PDF pages. Ensure page cleanup is called after each
page is
rendered and both PDF document and loading task are properly destroyed
with error
handling.