UTF-16 and -32 would require strings to include null bytes; this is
the case for all code points that correspond to ASCII. But a VCL
literal STRING by definition may not have a null byte (see vcl(3)).
So we couldn't have a pattern in a literal STRING, and that's the
show stopper. We also could not match regexen against literal strings;
that might not happen much in production, but it could in principle,
and the VMOD would have match against literal strings in the vtc tests.
So, maybe some day when we support matching against BLOBs (although
for that, we'd have to declare the pattern in a BLOB, which might
be just too odd).