A FontSelectPattern describes a general font request. It can be
used to find the best matching LogicalFontInstance. The instance
will be created based on a PhysicalFontFace, which is really a
factory since commit 8b700794.
Following this workflow, this moves the PhysicalFontFace pointer
to the instance and makes it constant.

Which leaves some special symbol font handling code in the hash
and instance lookup code path. It used to query the font face
directly from the instance.
I'm not sure of the correct handling. The related commits where
made to fix #i89002#, which has an attached test document.

1. commit 849f6182
2. commit 8c9823d3

The document is as broken as it was before the patch. The symbol
substitution still works, but the 'Q's are missing when displaying
a symbol font.

I also don't understand all the reinterpret_casts for fake font
ids. I guess this was used to prevent the crashes I see, where a
PhysicalFontFace referenced in a valid LogicalFontInstance is
freed and a later FontId check in the GlyphCache crashes. So this
now checks for a valid cache instead.

Change-Id: If8ee5a6288e66cfa4c419289fbdd5b5da128c6ea
Reviewed-on: https://gerrit.libreoffice.org/47279Tested-by: Jenkins <ci@libreoffice.org>
Reviewed-by: Khaled Hosny <khaledhosny@eglug.org>