Nesting !() extglob inside of first part of ?() extglob fails

[isaacs]

isaacs/node-glob#516

> new Minimatch('?(x-!(y)|z)').match('x-a')
false

> new Minimatch('?(x-!(y)|z)').set
[
  [
    /^(?=.)(?:(?!\.)x\-(?:(?!(?:y)|(?!\.z)?)[^/]*?)|(?!\.)z)?$/ {
      _glob: '?(x-!(y)|z)',
      _src: '(?=.)(?:(?!\\.)x\\-(?:(?!(?:y)|(?!\\.z)?)[^/]*?)|(?!\\.)z)?'
    }
  ]
]

[isaacs]

Curiously, this works, though:

> new Minimatch('?(x-!(y)|z)b').match('x-ab')
true

It's only failing when the containing extglob goes to the end of the path part.

[isaacs]

But not if the match for the !() also matches the rest of the pattern after the ?():

> m = new Minimatch('?(x-!(y)|z)b').match('x-bb')
false

[isaacs]

It's time for a proper recursive descent parser.

[isaacs]

I've got this almost working, and the new approach is a bit faster in some cases actually, since it can usually generate more efficient regular expressions. But it's a big refactor of the parser, so it may end up being a major version bump.