test-subset fails now because subset-face does not know how to compile itself.

Fixes https://github.com/harfbuzz/harfbuzz/issues/761
Fixes https://github.com/harfbuzz/harfbuzz/issues/762

It's a hb_face_t that has add_table() and in the future knows how to
compile itself into a font blob.

We keep a separate private NullPool in libharfbuzz-subset.so
This can fire back later since now there are two separate null objects,
and comparisons to the null pool can be confused.  Something to deal
with when it comes up. Ouch!

https://developer.gnome.org/glib/stable/glib-Testing.html#g-assert-cmpmem

I believe I'm happy with it now.

We expect the buffer to have no flags that the reference doesn't...
Meh. Makes MORX tests pass now. Need to better define the behavior.

Fixes https://github.com/harfbuzz/harfbuzz/issues/755

Fix build for subsetting code.