Implement reverse lookup instead of re-using next()

From https://github.com/behdad/harfbuzz/issues/161

Fixes https://github.com/behdad/harfbuzz/issues/161

https://github.com/behdad/harfbuzz/issues/161

https://github.com/behdad/harfbuzz/issues/139#issuecomment-154202645

The feature is enabled for any character in the Arabic shaper.
We should experiment with using it for Arabic subtending marks.
Though, that has a directionality problem as well, since those
are used with digits...

Fixes https://github.com/behdad/harfbuzz/issues/141

Fixes https://github.com/behdad/harfbuzz/issues/162

Unused currently.  To be used for Syriac stretch implementation.

https://github.com/behdad/harfbuzz/issues/141

This one is a hack, but should be ok.

I'm starting to really like how free these new scratch_flags are.

Hopefully makes clang build happy

Hopefully this wouldn't break any compiler.  There was some magic
moments to getting this to compile...

In anticipation of further changes.  No functional changes.

API change: all fonts created with hb_font_create() now inherit from
(ie have parent) hb_font_get_empty().

It makes the binary smaller AND faster.  Yumm!

Now that we have a buffer-wide scratch flags facility, use it to
optimize away a few passes.

Used to be 100.  If a tabe needs more than one or two edits, it's
probably completely bogus...

Might help with speeding up fuzzing for
https://github.com/behdad/harfbuzz/issues/157

Test passes now.

U+2011 is <noBreak> equivaent of U+2010, so we should do the fallback
for it.  Currently fails.