SSA CFG: Fix shuffler cycle caused by overeager pop of spilled top

[clonker]

• shrinkStack eagerly popped any spilled stack top before trying the swap logic below.
  When the spilled value was still demanded at an unfilled arg position, the pop just re-opened the demand, fixArgsSlot reloaded the same value, and the next shrink popped it again.
• Fix: the eager pop now only fires for junk and dead slots; a spilled top falls through to the swap logic, which places it directly into a demanded position (which the reload-side depth guards guarantee is swap-reachable). Since the reload only fires while an unfilled demand exists, and shrink swaps instead of popping in exactly that state, the pop and the reload can no longer alternate on the same state.
• Popping a spilled top stays possible, it's just demoted into the shrinkPriority cascade: canBeFreelyGenerated became slotCanBeLoadedOrPushed (meaning freely generatable or spilled), so a spilled slot scores as regenerable-out-of-position (4) or surplus (3) instead of being popped unconditionally first.

[blishko]

Hmm, does not seem ideal. It makes some cases worse. On the other hand, this seems to be a niche situation, so I am not against merging.

[clonker]

At least it doesnt regress any of the semantic tests :)

[blishko]

At least it doesnt regress any of the semantic tests :)

Probably because none of the semantic tests needs to spill? :)

[clonker]

Probably because none of the semantic tests needs to spill? :)

Some of them do. But it's admittedly not a super representative stat.