partition ASCII and non-ASCII byte classes when a Unicode word boundary is used

[BurntSushi]

From this issue BurntSushi/ripgrep#1513, it was discovered that the DFA would quit when the regex contains a Unicode word boundary even when the input was purely ASCII. It turns out that characters like | and { and } would get lumped into the same byte equivalence class as non-ASCII bytes, which would cause the DFA's non-ASCII circuit to trip and quit the DFA.

This should be easyish to fix. If the regex has a Unicode word boundary, then ensure that ASCII bytes are never lumped into the same equivalence class as non-ASCII bytes.

[BurntSushi]

This appears to have been fixed in #768. (It is also fixed in regex-automata.)