fix flate-encoded streams with PNG prediction; parse compressed xref streams
move stuff around and just use Biobuf* everywhere
main: print filename on error
remove stream/filter traces
make pdfeval update the pointer; add a dumb ref counter to objects
better api (less Pdf *pdf); eval more often; use null
add and use flate filter
rename some of the functions
attach dicts to streams if there is one
rewrite the API, support more object types and actual evaluation
fix xref parsing and add pdfeval to resolve indirect objects
add more object types, parse file trailer
pdfstring: octal chars