The six GPOS contextual subtable layouts are decoded twice, once to apply and once to report

[jakejackson1]

Follow-up to #90, which added the second copy.

Otl and OtlDump each decode the six GPOS contextual positioning subtable layouts, from
the spec, independently.

format	applied in Otl	reported in OtlDump
7/1	_applyGPOScontextPosFormat1, Otl.php:3930	reportGPOScontextPosFormat1, OtlDump.php:2563
7/2	Otl.php:3988	OtlDump.php:2620
7/3	Otl.php:4074	OtlDump.php:2677
8/1	Otl.php:4115	OtlDump.php:2711
8/2	Otl.php:4183	OtlDump.php:2764
8/3	Otl.php:4314	OtlDump.php:2821

Both walk the same records in the same order - a PosRuleSet chosen by coverage index, then
glyphCount, posCount, the input sequence one short of its count, then the
PosLookupRecords - and both had to get the one ordering trap right independently: Type 7
Format 3 puts posCount before the Coverage offsets where Type 8 Format 3 puts it after.

This is the same shape #112 was about, one level down. There the two copies had already
drifted; these two have not yet, because one of them is four weeks old.

Why it is not simply #112 again

The obvious fix - have the parser record what it read and the dump render it - does not
apply here, and #90 says why. TTFontFile never parses a GPOS subtable: it reads position
0's Coverage for the shaper's gate and nothing else. Making it parse all six layouts would
slow every font-cache build for the benefit of a debugging page.

So the sharing has to be at the level below both: the primitives, not the structures.

Where to look

OtlDump.php:2926-3010 already has them, extracted while #90 was written:

coverageOffsets()    the Coverage offsets of one sequence, read in place
coverageGlyphs()     follow them
posLookupRecords()   SequenceIndex + LookupListIndex pairs
glyphSequence()      a sequence naming one glyph per position
classSequence()      a sequence naming one class per position

Those are five methods that take a FontReader and return a structure, which is exactly
the shape of Mpdf\Fonts\Table\Coverage and Mpdf\Fonts\Table\ClassDef. Moving them there
and having both Otl and OtlDump call them would leave each side with its own reason for
reading - applying against the run, or rendering - and one copy of the layout.

Safety

tests/data/fontcache/*.json, tests/data/otldump/*.txt and tests/data/shaping/*.txt
between them pin both sides for every font in tests/data/ttf. Note that 7/2 is covered by
neither, for the reason in the issue filed alongside this one.