I didn't like it from the beginning.

Untested.

Fixes issue reported by Khaled Hosny with his Hussaini Nastaleeq font
and sequences like those added in the previous commit.

I know...

Filling in.

Patch and description from Jonathan Kew:

It turns out that some legacy Thai fonts provide OpenType substitution
features to implement mark positioning, but (incorrectly) put those
features/lookups under the 'latn' script tag instead of using 'thai' (or
possibly 'DFLT'). See
https://bugzilla.mozilla.org/show_bug.cgi?id=719366 for an example and
more detailed description.

Although this is really a font bug, I suggest that we could improve the
rendering of such fonts by looking for the 'latn' as a fallback if
neither the requested script nor "default" is found in
hb_ot_layout_table_choose_script. Suggested patch against harfbuzz
master is attached.

This does _not_ affect the other kind of legacy Thai font, where custom
code to support vendor-specific PUA codepoints would be needed. I'm not
keen to go down that path; IMO, such fonts should be ruthlessly stamped
out in favour of standards-based solutions. :)

JK

Since we already switched to accepting -1 as 'zero-terminated'.

hb_face_get_blob() renamed to hb_face_reference_blob(), returns a
reference now.

hb_face_[sg]et_index() added.

hb_face_set_upem() added.

Where causing issues for people with MSVC.

Buffer var allocation coming into shape

I've messed up a lot of stuff recently, different parts of the
shaping process are stumbling on eachother's toes because
manually tracking what's in which buffer var is hard.  I'm
going to add some internal API to track those such that mistakes
are discovered as soon as they are introduced.

Not used for anything right now.  Will use to get num_glyphs in the future.

Needed to be able to include <Windows.h> from hb-object-private.h.

That correctly reflects the reference ownership transfer happening.