ZJIT: Fix pass ordering in inliner

[tekknolagi]

Right now we do some loop like:

for i from 0 upto inline_max_iterations {
  type specialize
  ...
  inline
  ...
}

This is fine except that it means that anything inlined in the last round won't get type specialized. This is leading to more un-cached sends than without the inliner.

[tekknolagi]

We should probably do something like max_iterations+1 then conditionally inline in the loop