Should record types of interest be compositional?

If you do `if (x is int)`, we make `int` *and `int?`* types of interest for `x`.

If you do `if (pair is (int, int))`, should we also make `(int?, int?)`, `(int, int?)` and `(int?, int)` types of interest for `pair` (and the corresponding nullable record types)?

Or in general, if a record type is of interest, is nullability-variations on every field also of interest, including for the declared type.

The answer is probably "no", but maybe someone has a use-case where it makes sense, so putting the question out there.

For example:
```dart
void use((int, int) pair) { ... }

(int?, int?)? maybePair = ...; // Maybe nothing at every position, or at all.

if (maybePair == null) {
maybePair = (1, 2); // Assignment-promotes to (int?, int?) I assume, but not (int, int).
use(maybePair); // Won't work.
} else {
maybePair = (maybePair.$1 ?? 0, maybePair.$2 ?? 0); // No promotion from (int?, int?)
use(maybePair);
}
// ...
```

Start with the issue's examples of record types, nullable fields, and type promotion. No files or tests are named, so first determine the intended language rule and whether existing record-type interest behavior is documented elsewhere. Done means the nullability-composition question has a decided specification outcome and corresponding implementation scope.

Written by the indexing model from the issue text.