Allow non ascii characters in regexes

[vthib]

This is fixing the revert of behavior in YARA.

See VirusTotal/yara#1770 (comment)

Since YARA reverted the change due to existing rules containing such characters, we also want
to do the same. This is however not entirely trivial, for two reasons:

• The parsing library works on str, and changing this to instead work on &[u8] for the regex parsing is not easy, and i'm not even sure it is desirable.

• I wanted to add a new warning for such characters, as it can lead to unexpected failure to match properly. This required some rework to be able to generated warnings in code that did not previously.

This is now fixed properly:

• The parser now allows non ascii chars (except in classes, which also errors in YARA), and pass them in a new AST type.

• When we convert the regex AST to our internal HIR in boreal, we:

  • generate warnings when the new type is found
  • we properly generate the right HIR (which means some ugly code when handling the "" case, which only applies the repetition to the last byte of the char.

However, at the cost as some code a bit ugly in the AST->HIR conversion, we get the same behavior as YARA, and user friendly warnings on such cases. Overall, a nice change :)