xWidthAvg: Add `subset` support for non-latin character sets

xWidthAvg: Add subset support for non-latin character sets

Previously the xWidthAvg metric was calculated based on the character frequency as measured from English text only. This resulted in the xWidthAvg metric being incorrect for languages that use a different unicode subset range, e.g. Thai.

Supporting Thai now enables adding support for other unicode ranges in the future.

What's changed?

`@capsizecss/metrics`

The subsets field has been added to the metrics object, providing the xWidthAvg metric for each subset — calculated against the relevant character frequency data.

 {
   "familyName": "Abril Fatface",
   ...
+  "subsets": {
+    "latin": {
+      "xWidthAvg": 512
+    },
+    "thai": {
+      "xWidthAvg": 200
+    }
+  }
 }

There are no changes to any of the other existing metrics.

`@capsizecss/core`

Fallback font stacks can now be generated per subset, allowing the correct xWidthAvg metric to be used for the relevant subset.

The createFontStack API now accepts subset as an option:

const { fontFamily, fontFaces } = createFontStack(
  [lobster, arial],
  {
    subset: 'thai',
  },
);

seek-oss / capsize