C10m validation

[ezrakilty]

I don't know that we necessarily want to change around the benchmark like this, but the first commit does demonstrate the problem. I have to interleave two fibers to show that the local variables are overwritten between fibers, hence the extra complexity in the main loop. Probably a better way to get at that bug is to have a program more similar to pi that demonstrates it; of course in pi there's no distinguishability between the fibers, so we need to think.

Anyway, the latter commits fix two bugs:

• The return value of malloc was not checked
• The stack size of 2MiB was too big to allocate 10,000 of them simultaneously in the linear memory as given. Changing this size obviates the crash from malloc returning NULL. But maybe we want to make it configurable within the program instead?