This includes HB_DIRECTION_INVALID and HB_SCRIPT_INVALID.

The INVALID will cause a "guess whatever from the text" in hb_shape().
While it's not ideal, it works better than the previous defaults at
least (HB_DIRECTION_LTR and HB_SCRIPT_COMMON).

One should use hb_buffer_reset() really.

I really shall find a way to automate the test enumeration.  Otherwise
it's too easy to add a test and wrongly hook it up.  Did it twice today.

It's just sugar.

The problem with HB_TAG_STR() was that it expected a string of size 4
exactly, and unlike hb_tag_from_string() it doesn't pad the tag with
space characters.  So, the new name is more appropriate.

Uses the glib testing framework.  Wrote unit tests for hb-common.h
types.

Never hurts.

Since we now assert thos in hb-private.h, the int types cannot be wrong.
(Except for when someone else includes hb-common.h in a very broken
configuration, but that's not our problem!)

Plus, we don't use inline in the public headers, so remove that too.

That better matches OpenType spec.  Note that we enable it for all
Arabic-shaper scripts.  Ie. we enable it by default for Syriac too,
but the SyriacOT spec does not require it.  I think this is a more
useful compromise than special-casing for Arabic script alone.