This PR fixes a handful of oversights and behavioural discrepancies with the regular expressions used by our modeline strategy (detailed below).

Description

Whilst reviewing #4138, I recommended [ \t] be used instead of [:blank:]. However, I didn't think to check if the submitter applied the changes correctly. Only recently did I notice Vim's modeline regex had an erroneous [^\\[ \t]] (which should have been [^\\ \t] instead).

Realising there might be other things I missed, I recently took the time to double-check both regexes and verify their behaviour matches that of Emacs and Vim. Among the things I found were:

• Vim's modelines are case-sensitive.
  :help modeline says that modelines must be prefixed by “the string vi:, vim:, Vim: or ex:”. Mixed-case prefixes like vIM: or ViM: don't work, so VIM_MODELINE shouldn't be flagged as case-insensitive.
  
  
• Versioned modelines must start with Vim or vim.
  Our current pattern allows vi>600: ft=ruby and # ex>600: ft=ruby, neither of which actually work in Vim.
  
  
• Emacs's modeline pattern allowed newlines.
  This was due to \s being used instead of [ \t]. An embarrassing habit I've picked up from years of writing TextMate grammars is to assume \s only matches horizontal whitespace, instead of line-breaks as well. Ergo, something like this (obviously) shouldn't match:
  

  -*-
  markdown
  -*-
  

  
  
• Modeline mode wasn't disabled.
  So Ruby's insidious treatment of . meant that .*? also matched newlines, which it logically shouldn't. I resolved this by unsetting multiline mode with (?-m).
  
  
• Carriage returns were allowed inside vim: set … :.
  A trivial issue that probably won't ever surface in practice; fixed by updating classes to use \r\n instead of just \n.

(Template removed as it doesn't apply)