Fixes another 6% of the Khmer failures.

Brings down Khmer failures with Daun Penh font from 36% to 20%.

Khmer failures down from 58% to 36%.

Brings down Malayalam failures from 14% down to 3%.

No semantic change.

Almost...

No reordering yet.

Should not be any semantic change.  In preparation for implementing
pre-base reordering Ra.

This is a hack for now, to get us going with Khmer.  This will be
refactored properly later to move the complex logic into complex
shapers.

We are going to split matras without a Unicode decompositions in a way
that the second half takes the codepoint of the whole matra.  So,
position them where the second half is supposed to end up.

Apparently Uniscribe allows these, and they are used in some Khmer fonts
shipped with Windows, namely, Daun Penh.

So we can use them to switch default visibility to internal if desired,
and use these to make only declared symbols public.

Ouch, all our C++ inline functions are being exported (weakly) already.
Fix coming.

The calculations were wrong.

FreeType makes it really hard to set size and ppem independently.
For now, disable it.  Need to come up with a fix later.

This reverts commit 727135f3.

This is work-in-progress.  Didn't mean to push it out just yet.

Since we use the OpenType versions of Uniscribe functions, we are
relying on that version of the WINNT API.  Otherwise, usp10.h will hide
those symbols.

https://bugs.freedesktop.org/show_bug.cgi?id=50970