Add Stream Sieve benchmark from Typed Racket

[hugomg]

Closes #427.

This is a nice benchmark from the gradual typing literature but it has a problem... it hits the LUAI_MAXCCALLS limit due to too much recursion. I think I'll leave it here as a draft pull-request for now.

[hugomg]

There is a chance that this might stop blowing the stack if we implement tail recursion optimization (#211). It might also benefit from proper recursive types (#85), because there is one place where the as Stream cast makes the function no longer tail recursive.

[hugomg]

Another pre-requesite is optimizing indirect Pallene->Pallene calls (#354). Tail-call optimization is not possible for functions that cross a Pallene<->Lua boundary.

[hugomg]

Scratch that... It turns out that the benchmark isn't actually tail recursive. There are some function calls that are, but there are some that are not. In particular, the part in sift that happens before the % n == 0 check.

[srijan-paul]

Does this run in any of the regular lua interpreters?

[hugomg]

It works fine in Lua because the call stack limit is much higher (millions of calls, instead of hundreds of calls)