[fontscan] Better family substitutions

[benoitkugler]

This PR alters the way we perform family substitutions when looking for a fallback font, by applying them to the whole family list of the user query, so that substitution priorities are better respected.

Let me illustrate the change on the following example, where an user queries {CustomWeirdFont, Helvetica}, on a system with neither CustomWeirdFont nor Helvetica, and calls ResolveFace('a')

Before the change

No exact match occurs (step 1), so we apply substitutions (step 2). CustomWeirdFont is unknown from the library rules, so it is expanded by default to sans-serif for which the system has (many) fonts (with support for the rune 'a'). So ResolveFace returns the 'preferred' sans-serif face. It has not take into account the Helvetica substitutions.

After the change

We still look for an exact match, in the user query order. Like before, no one occurs. We now apply substitutions on the whole {CustomWeirdFont, Helvetica} list. Since the library has rules for Helvetica, the expanded families will look like {CustomWeirdFont, Helvetica, NimbusSans, ..., other sans-serif families}, with the main difference being that NimbusSans, which is the preferred substitution for Helvetica, has now higher priority than the default sans-serif substitutions coming from CustomWeirdFont.

This behavior is closer to what fontconfig does on Linux systems, and seems overall a better way of looking for fallbacks.

[andydotxyz]

Description sounds great and the code looks solid too, nice one :)

[whereswaldon]

Thank you; this is a definite improvement.