Users should #include <hb.h> (or hb-ft.h, hb-glib.h, etc), but
never things like hb-shape.h directly.  This makes it easier to
refactor headers later on without breaking compatibility.

For two reasons:

1. User can always call hb_buffer_pre_allocate() themselves, and

2. Now we do a pre_alloc in add_utfX anyway, so the total number of
reallocs is limited to a small number (~3) anyway.  This just makes the
API cleaner.

We need this to set data on objects safely without worrying that some
other thread unsets it by setting it at the same time.

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

hb_shape() now accepts a shaper_options and a shaper_list argument.
Both can be set to NULL to emulate previous API.  And in most situations
they are expected to be set to NULL.

hb_shape() also returns a boolean for now.  If shaper_list is NULL, the
return value can be ignored.

shaper_options is ignored for now, but otherwise it should be a
NULL-terminated list of strings.

shaper_list is a NULL-terminated list of strings.  Currently recognized
strings are "ot" for native OpenType Layout implementation, "uniscribe"
for the Uniscribe backend, and "fallback" for the non-complex backend
(that will be implemented shortly).  The fallback backend never fails.

The env var HB_SHAPER_LIST is also parsed and honored.  It's a
colon-separated list of shaper names.  The fallback shaper is invoked if
none of the env-listed shapers succeed.

New API hb_buffer_guess_properties() added.

This was a bizzare piece of API that I inherited from cairo.  It has
been wrong adding them to cairo in the first place.  Remove them before
someone uses them!

Returns the error status of the buffer.

The new name is self-documenting.

Return the length, whenever we return an array.  Makes it easier on the
language bindings.

One should use hb_buffer_reset() really.

Add variant integers to buffer item types.  More cleanup coming.

We expect buffer to be setup with default positions before GPOS.

Fix the second instance.