Generalize null-rejection checks into a reusable expression trait

### Is your feature request related to a problem or challenge?

[Discussion thread](https://github.com/apache/datafusion/pull/21549#pullrequestreview-4094422196)

Several optimizer rules need to determine whether an expression "rejects nulls" — i.e., returns NULL/false when one or more input columns are NULL. Today this logic lives in `EliminateOuterJoin`'s `extract_non_nullable_columns()` with explicit pattern matching for each expression type (comparison, IN, BETWEEN, LIKE, IS TRUE/FALSE, etc.). Every new expression type must be added manually.

### Describe the solution you'd like

Introduce a `NullRejection` trait on expressions, similar to Apache Calcite's [`Strong`](https://github.com/apache/calcite/blob/main/core/src/main/java/org/apache/calcite/plan/Strong.java#L100) class:

```rust

trait PhysicalExpr {
// ... existing methods ...

/// Returns whether this expression is guaranteed to be not-true
/// (i.e., NULL or false) when all given columns are NULL.
///
/// - `Some(true)`: definitely rejects nulls (safe to eliminate outer join)
/// - `Some(false)`: definitely does NOT reject nulls
/// - `None`: unknown (conservative default, assume not null-rejecting)
fn is_not_true(&self, all_null_cols: &[&Column]) -> Option {
None
}
}

```

Each expression type overrides with simple structural logic:

```rust
// Comparison (=, >, <, etc.): NULL on either side → NULL result
impl PhysicalExpr for BinaryExpr {
fn is_not_true(&self, all_null_cols: &[&Column]) -> Option {
match (self.left.is_not_true(all_null_cols), self.right.is_not_true(all_null_cols)) {
(Some(true), _) | (_, Some(true)) => Some(true),
(Some(false), Some(false)) => Some(false),
_ => None,
}
}
}

```

### Describe alternatives you've considered

_No response_

### Additional context

_No response_