Makes more sense, optimization-wise.

I know...

I see about 8% speedup with long strings with DejaVu Sans.

This is apparently what Uniscribe does.  Test case is:

  SEEN FATHA TEH ALEF

with Arabic Typesetting.  Originally reported by Khaled Hosny.

To be used for correct mark attachment to first component of a
MultipleSubst output.  That's what Uniscribe does.

Previously was failing to match fonts that didn't support CHARSET_ANSI.

There still remains a problem with the Uniscribe backend, in that if a
font with the same family name is installed, and is newer, the native
one is preferred over the font we provide.  Fixing it requires rewriting
the name table with a unique family name...

Introduced in 99c26957.
Broken mark-mark and mark-ligature stuff.