Where causing issues for people with MSVC.

This allows the Indic shaper to request decomposed characters.  This will
handle split matra for free.  Other shapers prefer precomposed
characters.

Nothing functional in there yet.

So far, we're parsing IndicSyllabicCategory.txt and IndicMatraCategory.txt
fils from Unicode Character Database and store them in an array to be used
by the shaper.  Also hooked up the shaper, but it does not do anything
right now.

Respectively, separate planner from the actual plan.

So we can liberally use the simple features of C++ that parts of the
codebase is already using.

This simplifies the code, reduces static data, and makes the design
more extensible to future additions of new scripts.

Add support for classic Mongolian script to the Arabic shaper.

Still work to be done around U+180E MONGOLIAN VOWEL SEPARATOR as it
should not be included in the final glyph stream the same way that
ZWNJ, etc should not appear in the final glyph stream.

But the joining part should be done.

There remains the question of how should the U+18A9 MONGOLIAN LETTER ALI
GALI DAGALGA be handled as it has General Category NSM but a letter
nonetheless.  For now, our generic logic makes this a joining T instead
of joining D as other Mongolian letters are.

We added Mandaic joining data to the Arabic shaper a while ago, but
were not actually using the Arabic shaper for Mandaic.  Fixed.

Not tested yet.

This is the object that a separate plan/execute shaping API will use in
between the two stages.

hb_ot_plan_t may be a better name, donno.

So we can apply it to class methods also.  Not sure if that works!

This reverts commit c939f6aff405ca7b10b1f1538f46148bff719fcb.