[early patch] Speed up decl chaining

[Rafael Espíndola]

>> Unbounded super linear algorithms in general provide a minefield that
>> is not very user friendly.
>
> I don't see how this fixes the superlinearity;

I fixes the decl chaining.

> it seems like it just moves
> it around.

It is possible. The patch is very incomplete. I simply changed the
order and started looking at what breaks.

>Doesn't it make getPreviousDecl linear? Some of the places you've
> changed from getPreviousDecl to getFirstDecl are also now linear.

Depends on which decl. Getting the first decl from the current one is
just one step, but it is fairly possible there is still superlinear
path I haven't seen yet

> The change to setObjectOfFriendDecl looks incorrect: we really wanted to
> look at the IDNS of the most recent decl, not of the first one.

Awesome. Do you have a testcase :-)

Thanks,
Rafael