Add `OrderedSet` to stdlib

[blhsing (Ben Hsing)]

@elis.byberi

The set.add() and set.remove() operations have an amortized worst-case time complexity of O(n). In other words, for large sets, they are no longer O(1).

That would only be the case if the set implementation didn’t resize the hashtable.

[Lucas_Malor]

(Excuse me for the spam, the previous reply had too much flaws…)

@rhettinger

• Would the implementation be a regular set augmented by a doubly-linked list? This is how collections.OrderedDict was implemented, making it slower and more memory intensive than regular dicts. Or would the implementation be a wrapper around the compact and ordered regular dictionary? The latter gives memory savings but is algorithmically poor for set applications with frequent deletions and reinsertions.

Excuse my ignorance, probably I’m asking a stupid question: can’t the existing algorithms of set be adapted to be used on a structure similar to the current compact dict structure? And why using a wrapper instead of a new struct?