Essentially move the glyph mapping to normalization process.
The effect on Devanagari is small (but observable).  Should be more
observable in simple text, like ASCII.

Based on patch from Philip Withnall.
https://bugs.freedesktop.org/show_bug.cgi?id=41095

Like Uniscribe does.

In this mode we try composing CCC=0 with CCC=0 characters.  Useful for
Hangul.

No logical difference so far.

In preparation for Hangul shaper.

Reported by Jonathan Kew.

Where causing issues for people with MSVC.

Number of failing shape-complex tests goes from 125 down to 94.

Next: Add Ra handling and it's fair to say we kinda support Indic :).

Fallback normalization is complete and working now!

Lets not worry about performance for now...