delete! in sorted containers incompatible with Dict and Set

[StephenVavasis]

The statement delete!(d,k), where d is a Dict or Set and k is a key, does not throw a key error even when k is missing from d. (In this case, it is a no-op.) On the other hand, SortedSet and SortedDict, throw a key error in this circumstance. Because these packages have been around for awhile, possibly someone is counting on this incompatible behavior, so I don't know whether to modify these containers to bring them in line with Dict. One subtlety: Dict and Set are not necessarily typed, whereas SortedDict and SortedSet are almost always typed. So one corner case is delete!(s,k) where s has keytype K, and k is not convertible to K.

[oxinabox]

From: #571 (comment)

Notice that there is an inconsistency with delete! that I didn't fix because I was afraid it would be a breaking change: #387

It is the general concensus that
turning errors into non-errors is not a breaking change.

[NHDaly]

One subtlety: Dict and Set are not necessarily typed, whereas SortedDict and SortedSet are almost always typed. So one corner case is delete!(s,k) where s has keytype K, and k is not convertible to K.

I would agree that this "strongly typed" key functions is a feature of SortedDict/SortedSet, and so I think we can keep this behavior while still claiming to implement Base.AbstractDict / AbstractSet. It seems reasonable to me to allow delete! nonexistent keys only when they're a subtype of K. 👍

[NHDaly]

Okay, i've opened #607, which does exactly this.