Fix a regression about fill rule (#131).

Address new CRAN check on the compiled code in sub-directories (#144, #146).

Partially support COLRv1 emoji fonts.

• COLRv1 emoji font is a color emoji, but not all color emoji is COLRv1
  format. For example, Noto Color Emoji has several variants, and it seems
  the primary one is CBDT/CBLC format.
• Additional information are currently just discarded.
  • The clip and layer composition information are just discarded. While
    this can be useful, it's not very easy to use these information in R.

Fix string2fill() and string2stroke(); when the second argument is a path
to a file, these unintentionally worked as string2path().

string2path() now generates the same outline as string2fill() and
string2stroke() (#69).

path_id and glyph_id are now 1-origin.

The result of string2fill() and string2stroke() now don't contain a
path_id column. I found the calculation of path_id had never been
correct, and it's probably better to remove it to avoid confusion.