AA implementation algo should be mentioned

[Frits van Bommel]

Frits van Bommel wrote:
[about inserts]
> IIRC the Phobos implementation uses a hash table with binary trees for 
> collision handling, which would mean /average/ O(1) for "good" toHash() 
> implementations, but worst-case O(log N) (for "bad" toHash()s).

Oh sorry, forgot about resizing. What does that make it, average 
amortized O(1)? :P