Reduce number of copies to lower memory footprint

The user bnncdv reported some regressions in 3.0.0 on the users mailing list.

If an input stream is used as source for a pdf every time the parser creates a view a copy of the RandomAccessReadBuffer holding the input stream is created. The duplicated byte buffers are reused but if it comes to larger pdfs the number of duplicated objects matter.

I was able to reproduce that behaviour.