The GSUB lookup list is walked twice, where rebasing the offsets would do

[jakejackson1]

Follow-up to #112.

TTFontFile::_getGSUBtables() walks the GSUB LookupList twice. The old code did too - the
second walk was introduced by a comment reading // Now repeats as original to get Substitution rules - and #117 kept the shape while giving both walks the same reader:

$GSLookup = $this->readLookupList($lookupListOffset, $gsubOffset, 7);   // TTFontFile.php:1274
$Lookup = $this->readGSUBsubstitutions($lookupListOffset);              // TTFontFile.php:1275

and inside that second one (TTFontFile.php:1334):

foreach ($this->readLookupList($lookupListOffset, 0, 7) as $i => $lookup) {

The only difference between the two calls is the third argument: the first asks for subtable
offsets relative to the start of GSUB, because the shaper reads a cached copy of GSUB alone
and has no idea where in the file it came from; the second asks for them relative to the
start of the file, because these passes are reading the file itself.

That difference is arithmetic, not parsing. The second walk re-reads every lookup header and
re-resolves every Type 7 Extension indirection - a seek and three reads per extension
subtable - to arrive at the same numbers plus $gsubOffset.

The fix is already written, in the subclass

OtlDump::absoluteSubtables() (OtlDump.php:1930) does exactly this rebasing in six lines,
for the GPOS side:

private function absoluteSubtables(array $lookups, $tableOffset)
{
    foreach ($lookups as $i => $lookup) {
        foreach ($lookup['Subtables'] as $c => $offset) {
            $lookups[$i]['Subtables'][$c] = $tableOffset + $offset;
        }
    }

    return $lookups;
}

Moving it onto TTFontFile and having readGSUBsubstitutions() take the $GSLookup and
$gsubOffset that _getGSUBtables() already holds removes the second walk. The two
structures are keyed differently - ['Subtables'][$c] against
['Subtable'][$c]['Offset'] - so the mapping has to be written, which is the only reason
this is an issue and not a one-line change.

Worth how much

Once per font-cache build, over the largest lookup list in the font: 273 lookups for Noto
Sans, more for a full CJK face. Not a hot path, and this is filed for tidiness rather than
speed - the comment at TTFontFile.php:1325-1330 currently explains the second walk as
inherent when the subclass three files away shows it is not.

Safety

tests/data/fontcache/*.json pins everything _getGSUBtables() produces for every font in
tests/data/ttf, and tests/data/otldump/*.txt pins what the dump makes of the same
lookups.