Synthesize productIterator for AnyVal case classes only

[joroKr21]

We can't reuse the inherited method from Product when the case class is AnyVal, because then the productIterator$extension would be missing from the companion. This restores binary compatibility with Scala 2.13.8 for AnyVal case classes.

Alternative to #10155

[som-snytt]

My comment on the other PR stands: it's a complication for no gain.

The alternative patch would be to generate only the value class extension method (for binary compatibility), which would just box and invoke the (mixin) method (instead of typedProductIterator). But that is also no practical gain.

My other joke was that the body could be just Iterator.single(value). That has the virtue of novelty.

[joroKr21]

Maybe I misunderstood your comment. I thought you'd be fine with it if I implemented it. I don't see it as much of a complication since we already treat AnyVal case classes differently. They gain is less code in general out there in the world which is good thing in my book. That hasn't changed since the original PR from Scala 2.13.9

[som-snytt]

Well, "fool me twice," "once bitten", etc. There must be an idiom for getting burned.

The same code is produced, it doesn't matter whether it's generated by mixin or case class. Both implementations just delegate to equivalent code.

After my PR is merged, to relieve my conscience, it's OK with me if Lukas likes this PR to further amend it.

[SethTisue]

Unless someone wants to make a convincing argument to the contrary very soon, let's stick with #10155 for 2.13.10, and then consider this one for 2.13.11.

[lrytz]

Given that this doesn't simplify case classes in bytecode (the method is still generated, just as a mixin forwarder instead of a synthethic case class member), and it doesn't simplify the compiler either, I think this change is not worh the (probably small) risk. Thanks @joroKr21!