split_by_number only splits Western Arabic Numerals and Full-Width Digits

[Numeri]

split_by_number is really nice for ensuring models learn to treat each digit separately, but the current implementation only applies to Western Arabic and Full-Width digits:

0123456789
０１２３４５６７８９

It'd be nice to include anything in the Unicode General Category Nd (Numeric, decimal digits), so that it covers Eastern Arabic numerals, Chinese numerals, etc.:

۰۱۲۳۴۵۶۷۸۹
零一二三四五六七八九十百千萬億
〇一二三四五六七八九十百千万亿

We could also cover the whole category N, which would also included non-digit numeric characters, like the ¹ (SUPERSCRIPT ONE) or Ⅹ (ROMAN NUMERAL TEN, not the letter X), but I'm not sure how intuitive it is to include those, or how often it even matters.