Add lazy stream loading to reduce memory for large PDFs

[EdwardBlair]

Adds ParsingOptions.LazyLoading (off by default) which defers reading stream bytes until they're actually needed. With this on, page.GetImages() gives you width, height, bounding box etc without loading the image data into memory - the bytes only get read if you actually access them.

For the file path overload this means using a FileStream instead of File.ReadAllBytes so the data stays on disk. Should help with #1171 where big images cause OOM on cheap nodes.

This could also work as a foundation for #980 - the lazy stream infrastructure means a future capabilities enum could skip processing images/paths entirely and the byte data would never leave disk.