Improve a check for inheritance relationship between types during binary operator re-inference in nullable walker

[jjonescz]

When re-inferring a binary operator in nullable walker, we previously replaced the original containing type with a new one if there was any implicit conversion between them, so e.g. operator==(ROS, ROS) would be re-inferred as a member of Span because there is a conversion from Span to ROS. But that doesn't mean the operator can be found on Span! Instead we should only allow conversions that make sense so the operator can be actually found on the re-inferred parent, e.g., a derived relationship (implicit reference conversion).

Fixes #80255.

[AlekseyTs]

I find the PR title "Narrow conversion kinds considered when reinferring methods by nullable walker" not very specific, I might even say confusing:

1. It looks like modified method affects only binary operators, not any methods
2. It is not clear what conversion specifically. Between which types, etc.

The description also doesn't clearly describe the context for the change, in my opinion. Consider adjusting. At least the title

[AlekseyTs]

Apart from title and description, LGTM (commit 1)

[jjonescz]

1. It looks like modified method affects only binary operators, not any methods

I think it does affect only binary operators though - the modified code is inside ReInferBinaryOperator.

2. It is not clear what conversion specifically. Between which types, etc.

I can update the title to say between derived/base types I guess. Thanks.

The description also doesn't clearly describe the context for the change, in my opinion. Consider adjusting.

I'm not sure what you mean, can you elaborate?

[jjonescz]

1. It looks like modified method affects only binary operators, not any methods

I think it does affect only binary operators though - the modified code is inside ReInferBinaryOperator.

Ah, that's what you are saying. Sorry, I misread it. I will update the title.