Confirm whether misused nullability attributes should produce diagnostics

[jcouv]

For instance, [EnsureNotNull] on a value type parameter, or [NotNullWhenFalse] on a parameter whose containing method doesn't return a boolean.

We need to review this, as it poses a compatibility concern.
My personal view is that if we use brand new attributes in the System.Runtime.CompilerServices namespace, then such warnings/errors are fair game (users are not supposed to define types in System).

Tagging @gafter

[gafter]

@jcouv What is to prevent users from compiling code that misuses those attributes that we provide and compiling with an old, or C++ compiler?

[jcouv]

@gafter Nothing. And if they do, the flow analysis for nullability will ignore such misused attributes.
Still, I believe we should help users use those new attributes properly.

[gafter]

But then the same code will not compile with new compilers - a breaking change.

[jcouv]

Closing due to Neal's point about breaking change.