Do not create conditional expression when holder and guard have the same type and guard overlaps with other branch

[phpstan-bot]

Summary

Fixes phpstan/phpstan#14469

When merging if/elseif scopes, createConditionalExpressions() could create a spurious conditional expression (CE) linking two variables that happened to have the same type in one branch. For example, in:

if ($var1) {
    $aa = $user->id === 10 ? 2 : null;
} elseif ($R['aa']) {
    $aa = $R['aa'];
}

In the elseif branch, both $aa and $R['aa'] have type mixed~falsy. The merge created a CE saying "when $aa is mixed~falsy, $R['aa'] is mixed~falsy". This CE later fired when if ($aa) narrowed $aa, incorrectly narrowing $R['aa'] to truthy — making !$R['aa'] a false positive "always false".

The fix adds a third skip condition in the inner loop of createConditionalExpressions(): when the expression does not exist in the other scope, the holder type equals the guard type, and the guard type overlaps with the other scope's guard type (isSuperTypeOf returns Maybe), the CE is skipped as unreliable. This correctly identifies cases where two variables were correlated only within a single branch, not causally linked.

Test plan

• Added regression test testBug14469 covering the original report and analogous patterns (if-constant-condition, falsey branch, with-else, multiple-elseif)
• Added NSRT test asserting $R['aa'] retains type mixed inside if ($aa)
• Verified existing regression tests pass: bug-14411-regression, bug-1306, full BooleanNotConstantConditionRuleTest
• Full test suite: 12131 tests, 0 failures
• make phpstan: no errors
• make cs-fix: no violations

[staabm]

(the diff got less readable with the last commit, but the source readability itself improved)

[staabm]

thanks!