Fix content data stream concatenation mangling output in cfFilterPDFToPDF

[sergio-gdr]

When running the following command on example.pdf
./pdftopdf 1 1 1 1 "" example.pdf > output.pdf

the resultant output.pdf is mangled.

This is better explained in this bug report.

It seems that the logic that provides the content data streams and concatenates them to form a single stream in an XObject is assuming a correct separation between the contents of successive streams.
To better illustrate: in example.pdf above (note that the sample pdf's have been run through qpdf --qdf --recompress-flate --compress-streams=n --object-streams=disable for illustration purposes), page 1's contents are

 /Contents [9 0 R 11 0 R 13 0 R 15 0 R]                                                                                                                                           

and looking at objects 9-11:

%% Contents for page 1
%% object 9
9 0 obj
<</Length 10 0 R>>
stream
q
endstream
endobj

%% object 10
10 0 obj
1
endobj

%% object 11
11 0 obj
<</Length 12 0 R>>
stream
q 0.1 0 0 0.1 0 0 cm
%% only the beginning of object 11 shown

This results in the following in output.pdf above:

%% resultant XObject. irrelevant info excluded
11 0 obj
<</Subtype /Form /Type /XObject /Length 12 0 R>>
stream
qq 0.1 0 0 0.1 0 0 cm

so the concatenation of streams 9 and 11 result in the (invalid) command 'qq', confusing pdf parsers and mangling the output.

With the patch applied, the output becomes

%% resultant XObject. irrelevant info excluded
11 0 obj
<</Subtype /Form /Type /XObject /Length 12 0 R>>
stream
q
q 0.1 0 0 0.1 0 0 cm

I'm not sure if this is the best solution for the problem, but hopefully the analysis can at least point to that.