The problem is that for anyone who knows how regexps work, this library provides an obfuscation layer. I personally prefer the free-spacing regexp support[1] found in a number of popular parsers. It lets you write a complicated regexp in small and/or indented chunks across multiple lines, including comments. This provides as much or more documentation than this "self-documenting" form and allows the full power of the regexp engine (which big regexps often seem to need). I've occasionally had to work out some pretty gnarly regexps, but when written in two-column code/comments form they end up being pretty understandable and maintainable.
[1] Example from the Ruby docs. IIRC, Python and a few other parsers have similar support: http://www.ruby-doc.org/core-1.9.3/Regexp.html#label-Free-Sp...
He's arguing that the nicer syntax is a valuable abstraction over regex's overly-complicated syntax, not that the syntax will be 100% the same as regex 100% of the time.
Apparently, we can't. So, you really want to improve your odds of writing correct code at the first attempt by making the language work with you rather than against.
Replacing a bunch punctuation with code that reads like prose is one way.
The fact you can't come up with a regex and be sure it's correct answers your first two questions.
Regexes are incredibly useful, but the syntax is just a bunch of noise. This library turns it into self-docummenting code, which is incredibly valuable. Novices might prefer showing their regex-fu, but after you've spent a good chunk of your life reading and writing code, you value clear code more.