attempt to offset vertical glyphs (#575) (#485)

[marklh9]

This patch would fix incorrect glyph position
highlighted in https://imgur.com/a/I0jJU as well
as in #485. I try to explain my point of view
and would like to know if I'm correct.

Note: for harfbuzz I have do some cairo translate.
In LibreOffice I don't. I don't know why.

Original issue is to fix text offset issue of
vertical writing in LibreOffice (#575)
after LibreOffice update its HarfBuzz from 1.3.3 to 1.4.8.
I tried to figure out how to use those values empirically.

Y_offset was 0 with HarfBuzz 1.3.3 and has
become -1160 (roughly 1EM upward) after upgrade to
1.4.8, I suspect that the offsets returned by Harfbuzz
are always repect to h_origin:

subtract_glyph_v_origin()
-> get_glyph_v_origin_with_fallback
-> guess_v_origin_minus_h_origin

But underlying renderer might expect positions relative
to the vertical origin. To compensate the result
I try to translate them back by calling
hb_font_add_glyph_origin_for_direction(),
to add the horizontal origin, hence make offset
relative to the v origin.

[behdad]

I'll try to review this soon. I've learned a lot about what's going on while debugging a similar issue on the Pango side. See #2765

[behdad]

I don't think this is correct. Our vertical support should be correct already.