GT4py.cartesian: For loop

[FlorianDeconinck]

Double for loop are a common pattern in column-physics (moist, radiation...). As of this writing the only way to emulate this behavior is to use the a while with a local variable which is convoluted and not user friendly.

Work was done at AI2 on the matter by Johann and Elynn, some of it remains in branches.

We could also limit the impact of the feature by allowing "sub-computation"on a dedicated axis e.g.

with computation(PARALLEL), interval(...)
  with sub-comptation(K, PARALLEL), interval(K, ...) or interval(..., K)
  with sub-comptation(DDIM[X], PARALLEL)

This would allow data-dimension parallelism (with restriction)

---

• Implement feature in gt4py
  OR
• Subtasks in finer grain tasks