Coverage, ClassDef, hb_set_t, and hb_map_t implement.

It was only used for C++<11 which does not allow default parameters
in function templates.  Looks like we cannot support <11 anyway, so,
start cleaning up.

The right condition to check for would have been "is_struct", which
we don't have.

Fix ambiguity of hb_sorted_array_t::item_t with gcc.  No idea if that's a gcc bug
or what spec requires, but using aliasing using seems to fix it.  It probably breaks
our non-C++11 bots, in which case I have to condition the change.  Testing.

Both, checks constexpr'ness, and fixes build with cra**y implementations
of assert() macro:

test-iter.cc:108:11: error: too many arguments provided to function-like macro invocation
  assert (hb_is_random_access_iterator (array_iter_t<int>));
          ^
./hb-iter.hh:186:42: note: expanded from macro 'hb_is_random_access_iterator'
  hb_is_random_access_iterator_of (Iter, typename Iter::item_t)
                                         ^
/usr/include/x86_64-linux-gnu/sys/cdefs.h:89:9: note: macro '__STRING' defined here
\#define __STRING(x)     #x
        ^
test-iter.cc:108:3: error: use of undeclared identifier '__STRING'
  assert (hb_is_random_access_iterator (array_iter_t<int>));
  ^
/usr/include/assert.h:91:21: note: expanded from macro 'assert'
   : __assert_fail (__STRING(expr), __FILE__, __LINE__, __ASSERT_FUNCTION))
                    ^

by partial-instantiating on Iter.

This reverts commit d6cbe96e2fc7bc8f1c10e631b52b1ef31ff9a6f5.

Isn't useful, as duplicate inheritance of same type results in ambiguity
errors...

Not enforcing it using type hierarchy.