I agree that /(a|b)+/ is better written as /[ab]+/ in the general case. But /(a|b)+/ was only intended to exemplify the difference between capturing and non-capturing groups in the split built-in function. Note that a and b in the (a|b) expression could represent any regex expression, not just the literal characters 'a' and 'b'. That's not true in a character class, which can only be composed of literal characters or another character class, e.g., \s or \w. See perlrecharclass.
... how that "?:" working in the regex.
The (?symbol(s) ...) syntax was introduced with Perl version 5.10 to support a multitude of regular expression extensions. The (? sequence was never valid in regexes prior to 5.10, so it was a convenient vehicle for these extensions. So you have
(?: non-capturing group)
(?> atomic group)
(?= positive look-ahead)
(?<! negative look-behind)
etc.
See Extended Patterns in perlre. See perlre and perlretut for info on the differences between capturing and non-capturing groups.
See also Special Backtracking Control Verbs for a similar syntactic twist: (* was never valid pre-5.10.
Give a man a fish: <%-{-{-{-<