At some point we should run tests under valgrind...

Hopefully fully fixes
https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=1856

This reverts commit 5daf3bd4.

If other atexit callbacks try to destruct the objects we destruct
in atexit callbacks, bad things will happen.

I'll come up with some other way to catch premature destruction
of HB-owned objects.

Fixes https://github.com/behdad/harfbuzz/issues/618

After 4a27c17e we do not need
the special GlyphID.cmp(hb_codepoint_t), so just make GlyphID
a typedef.

Fixes https://github.com/behdad/harfbuzz/issues/612

Patch from Jean Ghali on mailing list.

Fixes https://github.com/behdad/harfbuzz/issues/615

We'll see which bots I broke.

If offset is zero, we return Null() object.  Wasn't prepared for that.

Fixes https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=4088

We were using ft funcs because these are CFF fonts, but we can avoid
glyph names instead and keep testing both ot and ft.

See https://github.com/behdad/harfbuzz/pull/590#issuecomment-341194673

Fixes https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=4099

We don't need to test the actual flag values.  For unsafe-to-break,
the --verify in run-tests.sh checks what we are interested in.

Apparently calibri.ttf does this:
https://github.com/fonttools/fonttools/pull/1094#discussion_r148933791

Fixes https://github.com/behdad/harfbuzz/issues/607

Fixes https://github.com/behdad/harfbuzz/pull/605

Part of fixing https://github.com/behdad/harfbuzz/pull/605

Found using `codespell -i 3 -w -I '../harf-whitelist.txt'`
whereas the whitelist contents were:
```
beng
iff
pres
ot
te
teh
```

Seems to work.  Yay!

Still to do: run kerning if GPOS doesn't have 'kern' feature.