Move SPLIT_REDUCEOP into kernel.py ($500 bounty)

[George Hotz]

One LazyOp, two Programs.

Include a KERNEL in the middle of the LazyOp to get a second kernel.

1. Write a test with a hand coded LazyOp with a KERNEL in the middle. Get the lowerer generating the right thing.
2. Add support for 0-n ExecItems for each ScheduleItem.
3. Move SPLIT logic from lazy.py to kernel.py

Once it's in kernel.py, we can search over it.

[Emmanueltech]

Hi I would like to work on this bounty.