Add Map::insert_with_key

[killerswan]

As discussed in #2504, this adds an insert_with_key ([as in Haskell](http://hackage.haskell.org/packages/archive/containers/0.4.0.0/doc/html/Data-Map.html#g:5 as in Haskell)) to the Map trait.

[catamorphism]

Thanks! Can you resubmit without the "let's see if it compiles" comment? :-)

[nikomatsakis]

Can we give this another name? The name insert_with_key() makes no sense to me, every insert requires a key! Perhaps update() or insert_or_update()?

[nikomatsakis]

Sorry, I shouldn't have said 'makes no sense', sounds kind of harsh. I just meant that the name did not give me a hint at what the function did or how it might differ from any other insert(). In any case, seems like a useful operation!

[killerswan]

I'm about to push some fixes to this branch.

I've removed that comment about seeing if it works and stuck in a lower-level version which I expect to be faster. Now insert could be defined with insert_with_key, although I'm not sure we want to do that. Is rustc smart enough to optimize away something like |a,b,c| { c }?

I think I've clarified the tests a little bit, and I've also added an insert_with.

As far as names, how would update and update_with_key sound? I'm not that attached to the current names, but I've left them insert_with and insert_with_key like the Haskell versions.

[killerswan]

This push might still have a bug, but I have to go meet some people and don't have time to wait for tests to finish.

[killerswan]

OK, passing time CFG_DISABLE_VALGRIND=1 make check. :D

[brson]

Pushed as update and update_with_key. Thanks!