When we removed the separate Hangul shaper, the specific normalization
preference of Hangul was lost.  Fix that.  Also, the Thai shaper was
copied from Hangul, so had the fully-composed normalization behavior,
which was unnecessary.  So, fix that too.

Fixes 8 more Telugu tests.  Failures at 15 (0.00154548%).

Backport (forward-port?!) from upstream:

commit 3ab7b37bdebf0f8773493a1fee910b151c4de30f
Author: Behdad Esfahbod <behdad@behdad.org>
Date:   Mon Jul 30 10:50:22 2012 -0400

    Fix misc leaks

    https://bugs.freedesktop.org/show_bug.cgi?id=31992
    https://bugs.freedesktop.org/show_bug.cgi?id=31993
    https://bugs.freedesktop.org/show_bug.cgi?id=31994
    https://bugs.freedesktop.org/show_bug.cgi?id=31995

I didn't like it from the beginning.

The d1d69ec5 change broke Kannada badly,
since it was ligating consonants, pushing matra out, and then ligating
with the matra.  Adjust for that.  See comments.

This concludes the mark-attachment vs ligating interaction fixes (for now).

See comments.

Fixes https://bugzilla.gnome.org/show_bug.cgi?id=437633

Start squeezing more out of lig_id/lig_comp.

If two marks form a ligature, retain their previous lig_id, such that
the mark ligature can attach to ligature components...

Fixes https://bugzilla.gnome.org/show_bug.cgi?id=676343

In fact, I noticed that we should not let ligatures form between glyphs
coming from different components of a previous ligature.  For example,
if the sequence is: LAM,SHADDA,LAM,FATHA,HEH, the LAM,LAM,HEH form a
ligature, putting SHADDA and FATHA next to eachother.  However, it would
be wrong to ligate them.  Uniscribe has this bug also.

Unlike Uniscribe, hb-old returns glyphs in logical order, so the logic
does not need to duplicated for RTL.

Wasn't a huge fan of putting the burden on the user.  Just remove it and
do what we've got to do transparently.

This commit: a3313e54 broke MarkMarkPos
when one of the marks itself is a ligature.  That regressed 26 Tibetan
tests (up from zero!).  Fix that.  Tibetan back to zero.

And use it to speed up the hotspot by checking coverage directly in
the main loop, not 10 functions deep in.

Gives me a solid 20% boost with Indic test suite.  Less so for less
lookup-intensive scenarios.

Remove the "fast_path" hack from before.

Should give us some performance boost.

Shapers have a chance to reject a font in face shaper_data creation.
No need to allow failing during planning.