UM/MOD: Slightly improvements for entry and exit, more efficient loop structure.

[JeeeK]

First, the entry prevents the use of the stack pointer saved on the return-stack, in favour of pure register usage.
An early stack clean up saves duplicated code.
At last, the loop code is rearranged in that the 17th pass has been drawn into the entry section without the need for any undo measurements on exit. This simplifies the jumping cascade in the loop (and saves run-time too).

I'm sorry, but the modifications has not been tested yet. In the lack of an active development environment, I was not able to proof the correctness of these changes in run-time. The changes were made in accordance to the STM8 CPU programming manual.

[TG9541]

@JeeeK that's a nifty change, and since it's based on the PM alone it's all the more a feat! The automated test passes but I'll use @MrMarkB 's test scripts before merging. Thanks!