Fix result cache invalidation with @var referenced classes - #6343

Thanks! I was worried about the performance hit, so I instrumented the FileTypeMapper::getResolvedPhpDoc() cache while analysing phpstan-src src/ in a single process.

Good news: for Expression, Return_, Foreach_, If_ etc. your call hits the same cache entry that processStmtVarAnnotation() / getOverridingThrowPoints() already populated (same file/class/trait/function/text key). 100 % hits, CPU delta within noise (+0.3 %, 4 ABBA pairs).

Two redundancies though:

• Stmt\Property is already covered by the ClassPropertyNode branch below.
• InClassMethodNode, InClassNode, InFunctionNode and InTraitNode extend Node\Stmt and copy the original comments, so every method/class docblock is scanned twice.

Please restrict the new branch to the statements NodeScopeResolver::processStmtNode() feeds into processStmtVarAnnotation():

if (
	$node instanceof Node\Stmt
	&& !$node instanceof VirtualNode
	&& !$node instanceof Node\Stmt\ClassLike
	&& !$node instanceof Node\Stmt\ClassMethod
	&& !$node instanceof Node\Stmt\Function_
	&& !$node instanceof Node\Stmt\Property
	&& !$node instanceof Node\Stmt\ClassConst
	&& !$node instanceof Node\Stmt\Const_
) {
	$this->extractStmtVarTags($node, $scope, $dependenciesReflections);
}

I verified this variant: 251 calls, all cache hits, miss count identical to the branch without the PR, and your e2e test still passes (and fails without the fix).

Also please remove the str_contains('var') pre-filter. It's a bad heuristic (matches $var, "variable", "covariant"…) and with the restriction above the cache makes it unnecessary.

I like it, just run make cs-fix locally and forcepush. The other failures are moot (happening on other PRs too).

I'm sorry, there's now a conflict. I was fixing some result cache misses too.

Rebase the branch so that the history is linear and clean.

Thank you!