lint-commit-messages: count multibyte characters intelligently

[davidchambers]

If LC_CTYPE is not set appropriately, grep miscounts the characters in this commit message:

$ echo 'provide custom ‘inspect’ behaviour in Node without using ‘require’' | LC_CTYPE=POSIX grep "^.\\{74\\}$"
provide custom ‘inspect’ behaviour in Node without using ‘require’

The commit message actually contains 66 rather than 74 characters, if the multibyte characters are counted intelligently:

$ echo 'provide custom ‘inspect’ behaviour in Node without using ‘require’' | LC_CTYPE="$(locale -a | grep '^C[.]UTF-8$' || echo "$LC_CTYPE")" grep "^.\\{66\\}$"
provide custom ‘inspect’ behaviour in Node without using ‘require’

The fix provided in this pull request is to specify LC_CTYPE=C.UTF-8 if C.UTF-8 is one of the available locales.

Another solution would be to encourage users to set LC_CTYPE locally and on their continuous integration servers, but I think we should make an effort to count characters correctly no matter how the system is configured.