To fix older compilers again (this was the case in hb_array_t).

hb-ot-layout-common.hh:1353: note: candidate 2: operator[](T*, int) <built-in>
hb-ot-layout-common.hh:1354: error: ISO C++ says that these are ambiguous, even though the worst conversion for the first is better than the worst conversion for the second:
hb-iter.hh:63: note: candidate 1: Item& hb_iter_t<Iter, Item>::operator[](unsigned int) const [with Iter = hb_array_t<const OT::IntType<short unsigned int, 2u> >, Item = const OT::IntType<short unsigned int, 2u>]
hb-ot-layout-common.hh:1354: note: candidate 2: operator[](T*, int) <built-in>
hb-ot-layout-common.hh: In member function 'bool OT::ClassDef::serialize(hb_serialize_context_t*, hb_array_t<const OT::IntType<short unsigned int, 2u> >, hb_array_t<const OT::IntType<short unsigned int, 2u> >)':
hb-ot-layout-common.hh:1490: error: ISO C++ says that these are ambiguous, even though the worst conversion for the first is better than the worst conversion for the second:
hb-iter.hh:63: note: candidate 1: Item& hb_iter_t<Iter, Item>::operator[](unsigned int) const [with Iter = hb_array_t<const OT::IntType<short unsigned int, 2u> >, Item = const OT::IntType<short unsigned int, 2u>]
hb-ot-layout-common.hh:1490: note: candidate 2: operator[](T*, int) <built-in>

hb_array_t is its own iterator...

Should only be done by subclass.

Starting to get the hang of when take which.

Fixes https://github.com/harfbuzz/harfbuzz/issues/1502

  C:\projects\harfbuzz\src\hb-uniscribe.cc(709): error C2666: 'hb_vector_t<OPENTYPE_FEATURE_RECORD,8>::operator +': 3 overloads have similar conversions [C:\projects\harfbuzz\build\harfbuzz.vcxproj]

I'm still going back and force...

This reverts commit 969ff3c7.

Not sure if we are going to use it.  But might incentivize us to.