[DISCUSSION] Adding .skb.iter_eval?

[jeromedockes]

I wonder if it would be useful to allow evaluating an expression step by step, to inspect the results, debug, build reports, test etc. We could add a method .skb.iter_eval() that returns a generator which yields each sub-expression and its result as it evaluates them

>>> import skrub

>>> a = skrub.var('a')
>>> b = skrub.var('b')
>>> c = a * a + b

>>> for item in c.skb.iter_eval({'a': 10, 'b': 20}):
...     print(item)
(<Var 'a'>, 10)
(<BinOp: mul>, 100)
(<Var 'b'>, 20)
(<BinOp: add>, 120)

(note as the pr shows in terms of implementation there is almost nothing to do, as we already use a loop and not recursion to evaluate the expressions)

[glemaitre]

When @jeromedockes presented this feature, I thought about some discussions that we had in the past quite related (@GaelVaroquaux): being able to have a sort of probe in the graph for instance to get the feature importance. I'm wondering if this way of iterating would be useful to go in this direction.