[PATCH] Fix a performance problem in gep(gep ...) merging

[Daniel Berlin]

>
> I suspect not often, but I don't really know. It might be a good idea to add
> a statistic to count how often we hit the limit. To be clear, I'm not
> against this change, but I wanted people to keep these recursion depth
> limits in mind as it might be relevant to adjusting for any regressions
> should they emerge.
>

+1.
It would also be good to get an understanding how much, if any, of
this walking we might eventually want to move out of BasicAA (which is
stateless), to something that is more stateful and invalidates
occasionally.