GitLab

This brings the lcms2 code path and the non-lcms2 code path closer to each
other in terms of color reproduction.
So far the following points were missing, which are now added in this commit:
- Both code paths did either not adjust for a different source white point at all (lcms2 code path)
  or did some simple scaling. The code has now been adjusted to use the Bradford transform
  to adapt to either the D50 or D65 white point, depending on the code path.
- The non-lcms2 code path so far used the square root as an approximate gamma function.
  The correct sRGB gamma function has now been added.

vsnprintf actually works fine than qvsnprintf on MSVC nowadays so use
that

Initialize maxUsedGlyph after the early check returns

oss-fuzz/29629

oss-fuzz/23086

oss-fuzz/8795

Make sure colorspace doesn't need more comps than we have

oss-fuzz/29522

oss-fuzz/29460

oss-fuzz/29386

Otherwise we will end up doing an uninitialized memory read down the
road

oss-fuzz/10040

oss-fuzz/11744

oss-fuzz/11197

oss-fuzz/10059

oss-fuzz/29335

oss-fuzz/29330

oss-fuzz/29305

oss-fuzz/27783

oss-fuzz/29269

It seems we already did this check a few lines above, and indeed we did,
but on very broken documents, if arr[1] is a Ref, getting objects may
end up in a reconstruct xref call which may end up changing the type of
arr[1] the next time we ask for it

oss-fuzz/29260

oss-fuzz/29217

oss-fuzz/29241

Doesn't seem to regress any of the valid files i have and saves some
broken ones that loop for hours

Fixes crash in broken files #1022

oss-fuzz/29201

This makes the memory handling more error-proof, because
all necessary calls to 'delete' happen automatically.

This patch introduces several calls to std::unique_ptr::release,
and hence break the explicit ownership chain established
by using std::unique_ptr.  Not all of the calls are strictly
necessary, but without them the patch would get a lot
bigger than it is.  Better keep that for another day.

Now that we have oss-fuzz coverage i guess more will be coming soon ^_^

oss-fuzz/29199

It's unneeded and can be relatively easily used to create stack
overflows

oss-fuzz/29184

None of them is problematic, but being more strict let's use enable the
goostring-format-checker clang plugin