#20 breaks subclassing jQuery

[laughinghan]

Unclear whether this is worth "fixing" (whatever that means), but I think it's worth at least discussing even if we decide just to document a workaround.

If you subclass jQuery with Pjs, many methods that aren't supposed to mutate this, like .appendTo(), will add elements to this: http://jsbin.com/faxifed/edit?html,console,output

This was causing exponentially growing jQuery sets in MathQuill (https://github.com/mathquill/mathquill/pull/629).

This is because jQuery internally calls this.constructor() which normally creates a whole new jQuery object, but for Pjs subclasses starting with #20 on v5.0.0 just returns this: https://github.com/jquery/jquery/blob/305f193aa57014dc7d8fa0739a3fefd47166cd44/src/core.js#L72-L84

Thanks @jwmerrill!

[laughinghan]

Seems like the main reason jQuery can be subclassed "idiomatically" (i.e. the reason Coffeescript and ES6 classes don't break here) is because they allow constructors to return objects: http://jsbin.com/qinare/edit?html,console,output

However, they're not "well-behaved" subclasses, since if you try to instantiate them you get a jQuery object that isn't an instance of the subclass.