And use it in FontInfoScanner::scanFonts

Fixes #752

And we already have a resObj down the function that is the Resources
object so that should be actually called resObj

In this case we have to be a bit more careful since we're changing code that
used to assign to 0,0 and now INVALID is -1, -1 but i'm confident it's
fine

inReplyTo seems to be only used in qt5/src/poppler-annotation.cc and
i've updated the code to use the boolean isInReplyTo instead of checking
the ref number directly

The change in Dict only affects its two callers, one in Annot and one in
Catalog. The one in catalog has been updated, the one in Annot doesn't
seem to need updating (and moreover if you check history before using
the new Dict function was using -1, -1 as ref not initialized)

The change in Array only affects its one caller, in Function, whose code
has been updated

The embFontID change is something that was forgotten in the previous
commit about using Ref::INVALID

The change for iccProfileStreamA is only local to that function and has
been changed to use Ref::INVALID in all its uses

Lookup for PrintPageRange viewer preference
in constructor of ViewerPreferences and make
it available via getPrintPageRange() method.

poppler/poppler#290

Lookup for NumCopies viewer preference
in constructor of ViewerPreferences and make
it available via getNumCopies() method.

poppler/poppler#290

Lookup for PickTrayByPDFSize viewer preference
in constructor of ViewerPreferences and make
it available via getPickTrayByPDFSize() method.

poppler/poppler#290

Make sure the index of allIntersections we access is valid

Fixes #748

QStringliteral+MSVC+non ascii chars doesn't play very well

Windows.h defines max/min macros, these extra parentheses makes it happy
enough ...

introduced by commit 86326030

The function to detect 7bit Ascii assumed the check was being
done on a plain C char type (1 byte length) whereas the passed
Unicode type is bigger (defined as unsigned int) and so can hold
larger values. So fix our detection for an Ascii7 char appropriately.

Fixes issue #743

https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=8443

oss-fuzz/8631

Since a Parser cannot be used without a Lexer, make the Parser own the Lexer by value and construct it in place.

When a radiobutton (belonging to a normal radiobutton group)
has a /V key matching his 'OnStr' state, then when you
ask that radiobutton for his state (eg. radiobutton->state())
it will wrongly return 'true', when really the active
radiobutton is another one in the group.

This happens because the faulty radiobutton was not passing
the getState() call to his Parent (which every radiobutton
in a group should do, as the Parent stores the value of the current
active item).

The code was not doing it because it had a valid AppearanceState
(/V key). That behaviour may be right for checkboxes but not for
radiobuttons.

A testcase is included. An example definition of an
affected radiobutton follows:

/F 4
/FT /Btn
/Ff 49152
/AP /N /Beer 59 0 R /Off 61 0 R
/AS /Beer
/MK /BC [1,0,0] /BG [1,1,1] /CA (4)
/P 20 0 R
/Parent 8 0 R
/Rect [235.277,654.247,249.224,668.194]
/Subtype /Widget
/Type /Annot
/V /Beer

Fixes issue #159

All the callers already have a ref so makes no sense to unbox it

They are not supported in the spec but Adobe supports them and there's
	files outthere with these kind of strings, so we're supporting them

by using QString::reserve it should be faster and use less memory

They are not part of the standard but Adobe seems to support them and
there's files out there like that so better to support them than not

I can produce such annotations when adding annotations to a PDF
attachement from the standard mail app on my iPhone (iOS 12.1).
They currently all show as "ÿþÚ" rather than the actual string content.

UTF16-BE vs UTF16-LE is detected by inferring the endianness from the
first two bytes of the string (0xFF 0xFE and 0xFE 0xFF aka Byte Order
Marker).

Pass address of the first byte after end of the source buffer
to downsample_row_box_filter() so that we can check
that we don't run out of it.

Fixes issue #736

Use double precision for computation of coverage
of the left most pixel in the box filter.

Issue #736

The old code was not updating the parent member on setParent

Closes #742