so in sync with fonttools behavior

hb-coretext.cc: clean up macosx test

TARGET_OS_OSX was introduced only in late OS versions
so always returns as "0" on older systems.

if !TARGET_OS_IPHONE can work, as it returns as !0 on older
systems where TARGET_OS_IPHONE is not defined, but is not
specific

if TARGET_OS_MAC && !(defined(TARGET_OS_IPHONE) && TARGET_OS_IPHONE)
is both specific and accurate on all systems.

and CFF retain-gids fixes

along with api tests & expected results

Fixes https://github.com/harfbuzz/harfbuzz/issues/1535

[subset] Update the subset fuzzer to get  options to use from test case.

[subset] Update the subset fuzzer to determine which options to use based on data in the fuzzing test case.
Add support for toggling retain_gids.

[subset] Add --retain-gids option to the subsetter.

[subset] Re-add glyphs array to subset plan, with new name 'glyphs_deprecated'. Switch CFF subsetting to use it.

Re-implement HMTX/VMTX subsetting. Update it to support glyph renumbering and simplify the implementation.

Update maxp to use the correct num glyphs.

- Extract code out into helper methods in several places.
- Bundle loca address, size and is short into a struct.

[skip ci]

[skip ci]