Improve isset specification in falsy scope

[herndlm]

Continues specifying the type for a variable in a falsy isset if it exists in the scope.

This affectively leads to a more refined variable type and "fixes" type narrowing with empty() in a truthy scope too (see TypeSpecifierTest changes). The reason for that is that e.g. empty($foo) translates to !isset($foo) || !$foo and without types for $foo on the left side of the BooleanOr SpecifiedTypes::intersectWith() would simply drop the type info for the right side.

[herndlm]

Both the composer and doctrine failure seem legit.

Composer deals with arrays in negated contexts where only booleans are allowed, see https://github.com/composer/composer/blob/9a69c12a0799d5a974f5ec5d2e78be3495b0b06c/src/Composer/Command/DiagnoseCommand.php#L725

Doctrine checks the $className with empty which is unsafe since it can still be an empty string, see https://github.com/doctrine/orm/blob/2.13.1/lib/Doctrine/ORM/Mapping/ClassMetadataInfo.php#L3763

[herndlm]

I figured out how to do it more generically with the help of MutatingScope::issetCheck() for now.

Unfortunately I had to revert a tiny change from my previous #1784. It was not there until a couple hours ago still anyways :P The reason is that it will mess with the falsy scope by not unsetting $moreSpecificTypes for the ArrayDimFetch (which is set to make the expression inside isset() nullable I believe) and then MutatingScope::isSpecified() wrongfully returns true in the wrong context. Related to https://github.com/phpstan/phpstan-src/pull/1307/files#diff-90699c8d7b353d8df3308429a98104b255870f8411c204811b169f528535b022R91 where @rajyan introduced this and hinted it in the description too I think. I think it's fine to keep it like that for now, but this should be checked and cleaned-up
That was also one of the reasons why I did not succeed in MutatingScope / IssetCheck duplication-removal yet :/

But anyways, I think this PR improves falsy isset() and truthy empty() massively :)

[herndlm]

regarding this - here SpecifiedTypes::normalize() wouldn't even be needed and then it would work.. I think I can inline the normalization into SpecifiedTypes::intersectWith() and make it a bit smarter. I'd look at that in a dedicated PR.
ah I remember what the ugly parts here were - the Scope that needs to be passed to intersect then or the other type that needs to be passed to normalize. ugh

then the simplest working thing would be to use the correct type, e.g. the following on MutatingScope to use it inside SpecifiedTypes:

        public function getTypeOrNativeType(Expr $node): Type
	{
		return $this->treatPhpDocTypesAsCertain
			? $this->getNativeType($node)
			: $this->getType($node);
	}

or also pass around $treatPhpDocTypesAsCertain 🤔
but IIIRC there were/are other PRs or issues that touch the same subject

[ondrejmirtes]

I don't think this is right. Native type has to be evaluated alongside the full type. We're always interested in both at the same time. That means that TypeSpecifier and SpecifiedTypes should carry both at the same time.

[herndlm]

I'm gonna try further looking into this since it bothers me quite a bit

[ondrejmirtes]

I think it's a stepping stone to implement my suggestions here: #1627 (comment)

[ondrejmirtes]

Thank you!