Fix false positive nullsafe.neverNull on left side of ??

[kemo]

Problem

IssetCheck unconditionally reports ?-> property access as unnecessary (nullsafe.neverNull) when it appears on the left side of ??, isset(), or empty().

This is correct for isset()/empty() — PHP handles null objects natively in those constructs without throwing. But for ??, it is a false positive: the null coalescing operator does not catch TypeError from property access on a null object.

class Foo {
    public function getBar(): ?Bar { /* ... */ }
}

// PHPStan reports: Using nullsafe property access "?->baz" on left side of ?? is unnecessary.
// But removing ?-> would cause TypeError at runtime when getBar() returns null:
$foo->getBar()?->baz ?? 'default';
//              ^^^
//  Without ?->: null->baz throws TypeError (not caught by ??)
//  With ?->:    null?->baz evaluates to null, ?? returns 'default'

Root cause

In IssetCheck::check(), the NullsafePropertyFetch block (line 254) fires unconditionally regardless of context. It doesn't distinguish between ?? (where ?-> prevents TypeError) and isset()/empty() (where PHP already handles null objects).

Fix

When the identifier is nullCoalesce, check whether the object being accessed ($expr->var) has a nullable type. If it does, the nullsafe operator is genuinely needed and no error should be reported.

isset() and empty() behavior is unchanged — ?-> remains correctly flagged as unnecessary there since PHP handles null objects natively in those constructs.

Test changes

• Removed 3 false positive expectations from NullCoalesceRuleTest::testBug7109() (lines 17, 28, 66 of the test data file)
• Added new regression test testNullsafeCoalesceNullableObject() covering:
  • Chained nullable method calls with ?-> on left of ?? (no error expected)
  • Single nullable return with ?-> (no error expected)
  • Non-nullable object with ?-> (correctly reports "Expression not nullable")

[VincentLanglet]

Please provides a php snippet with leads to an error.

[VincentLanglet]

I have no error with
https://3v4l.org/5TbIf#v8.3.30

so the ?-> seems useless.

[kemo]

Closing this — after testing actual PHP behavior, I confirmed that ?? does handle null->property access via FETCH_OBJ_IS opcode, so the nullsafe on property access is genuinely unnecessary on the left side of ??. The distinction is that null->method() throws (method calls always execute), but null->prop is handled by ??. PHPStan's current behavior is correct. Apologies for the noise.