Add/test training against more ligature forms

[ryanfb]

Passing --ligatures to text2image gives us st ligatures in Cardo and EB Garamond, as well as ffi ffl (more?) in Garamond. If we could find a way to conditionally turn on the hlig OpenType font feature in Garamond we should be able to get ct ligatures out of it (and maybe Cardo?).

Wyld has ligatures mapped into the ASCII characters:
ÌËÊÉÈÇÅÄÃÂÁÀ
So we'd need to run text2image against these then substitute them back to the characters we want before running training.

[ryanfb]

For hlig support, it seems like we would need to modify Tesseract's text2image to support a new e.g. --opentype_features argument that could call pango_ot_ruleset_add_feature with the corresponding PangoOTTags (like Pango's syriac-fc.c does).