tbh I did not yet understand why this PR works, but I wonder whether we could use the same logic to also infer a better result when the iteratee is a known constant array?

like in

• Help phpstan understand what shape array has been created when looping over an array with known shape phpstan#13000
• False positive "Offset does not exist" when populating with foreach phpstan#2294

Why this works - I look at $a[$k] type at the end of the loop and at each continue point and if some other conditions are true (like there are zero break points), I rewrite the whole array value type with the $a[$k] type unioned from all the points I looked at.

When $foreach->expr is a constant array I feel like we need a different approach, possibly analysing the foreach separately for each item (but run $nodeCallback for the rules still only once).

Also I want to look at foreach ($a as &$v) to understand it better that also needs a different approach.