[HarfBuzz] Further splitting libharfbuzz.so

[Cosmin Apreutesei]

>   * hb-ft is a very small piece; convenience code you can even call it.  Most
> "serious" users of harfbuzz don't use hb-ft, but small users all use it.

I see pango uses ft too to implement the callbacks for
hb_font_set_funcs(). Are there other harfbuzz users that don't use ft
to implement those callbacks? I don't mind the ft dependency, I just
want to look at more examples of how harfbuzz is used so I can
understand it better.

>     - As the default Unicode callback provider.  If glib is disabled, harfbuzz
> builds an internal piece of code called UCDN that provides the same data.  It
> results in an extra 100kb of binary as well as duplicated data.  I like to
> avoid that, but don't want it to affect my ABI.

I have the opposite use case - I would like to avoid the much bigger
glib dependency. I am curious why hb_unicode_funcs_get_default()
chooses icu and glib with priority over ucdn when they are both
available. Are there any downsides (other than the 100kb) to using it
in place of icu or glib?